{"id":16724,"date":"2026-07-22T09:20:27","date_gmt":"2026-07-22T13:20:27","guid":{"rendered":"https:\/\/wp.glbgpt.com\/?p=16724"},"modified":"2026-07-22T09:20:27","modified_gmt":"2026-07-22T13:20:27","slug":"gemini-3-6-flash-review","status":"publish","type":"post","link":"https:\/\/wp.glbgpt.com\/zh-hk\/hub\/gemini-3-6-flash-review","title":{"rendered":"Gemini 3.6 \u5feb\u901f\u8a55\u6e2c\uff1a\u5b9a\u50f9\u3001\u57fa\u6e96\u6e2c\u8a66\u3001API \u8207\u6548\u80fd"},"content":{"rendered":"<aside class=\"g36-review g36-definition\" aria-label=\"Gemini 3.6 Flash at a glance\" style=\"background:linear-gradient(135deg,#f7fbff,#edf4ff);border:1px solid #bfd3f4;border-left:6px solid #2459bd;border-radius:14px;box-shadow:0 10px 28px rgba(18,38,82,.08);color:#26395f;margin:26px 0;padding:18px 22px\"><span style=\"background:#dbeafe;border-radius:999px;color:#194584;display:inline-block;font-size:11px;font-weight:850;letter-spacing:.12em;margin-bottom:10px;padding:6px 11px;text-transform:uppercase\">\u4e00\u89bd\u7121\u907a<\/span><p style=\"margin:0\"><strong>Gemini 3.6 \u9583\u5149\u71c8<\/strong> is Google&#8217;s stable, natively multimodal workhorse model for text generation, coding, document analysis, and tool-using agents. It accepts text, images, audio, video, and PDFs, supports a 1,048,576-token input window, and produces text or tool calls rather than native image or audio output.<\/p><\/aside>\n\n\n\n<p class=\"g36-review g36-review__dek wp-block-paragraph\"><strong>Gemini 3.6 Flash is worth testing<\/strong> for bounded coding, multimodal analysis, and supervised agent workflows. This Gemini 3.6 Flash review finds a credible upgrade over 3.5 Flash: Google lists lower output pricing and stronger results on several agentic benchmarks, while our first-party tests showed exact structured extraction, chart reading, and focused code repair after sensible output limits. It is not a universal winner. Our synthetic 128K retrieval task failed twice, and the model&#8217;s local browser plan invented a hidden element ID even though it identified the visible target correctly.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The practical recommendation is to pilot Gemini 3.6 Flash behind workload-specific acceptance tests, measure total cost per successful task, and keep human confirmation around irreversible actions. The official numbers, our Broly gateway-inclusive measurements, third-party benchmarks, and early social reports below are labeled separately so that unlike evidence is not blended into one score.<\/p>\n\n\n\n<div class=\"g36-review g36-verdict\" aria-label=\"Review verdict\" style=\"background:linear-gradient(135deg,#122b5b 0%,#2159a8 58%,#674db1 100%);border:1px solid #365d9b;border-radius:20px;box-shadow:0 20px 48px rgba(21,42,92,.20);color:#f8faff;display:grid;gap:14px 18px;margin:28px 0 34px;padding:28px\"><span class=\"g36-verdict__eyebrow\" style=\"color:#cfe0ff;font-size:12px;font-weight:800;grid-column:1\/-1;letter-spacing:.14em;text-transform:uppercase\">\u7c21\u8a55<\/span><span class=\"g36-verdict__label\" style=\"background:rgba(255,255,255,.14);border:1px solid rgba(255,255,255,.28);border-radius:999px;color:#fff;font-size:22px;font-weight:800;line-height:1.2;padding:14px 18px\">Recommended with workload testing<\/span><span class=\"g36-verdict__copy\" style=\"color:#f3f7ff;font-size:17px;line-height:1.55\"><strong>\u6700\u9069\u5408<\/strong> supervised, testable agent work where multimodal input and stronger planning justify more cost than Flash-Lite.<\/span><span class=\"g36-verdict__caution\" style=\"border-top:1px solid rgba(255,255,255,.25);color:#e7edff;grid-column:1\/-1;padding-top:14px\"><strong>Watch:<\/strong> long-context retrieval, output-token caps, verification claims, and gateway-specific feature gaps.<\/span><\/div>\n\n\n<p><!-- Google \u7039\u6a3b\u67df\u6fb6\u682d\u647c\u9365\u5267\u5896\u951b\u5c7e\u68e4\u95c7\u20ac\u6d93\u5a41\u7d36\u951b\u6cb4rc=https:\/\/storage.googleapis.com\/gweb-uniblog-publish-prod\/images\/gemini-3-5_3-6_3-5-Cyber__key-art__statement_.width-1300.jpg --><\/p>\n\n\n<h2 id=\"quick-verdict\" class=\"wp-block-heading g36-review\">Gemini 3.6 Flash review: pros and cons<\/h2>\n\n\n\n<figure class=\"g36-review wp-block-table g36-table g36-pros-cons\" tabindex=\"0\" role=\"region\" aria-label=\"Gemini 3.6 Flash review pros and cons\" style=\"border:1px solid #d9e1ef;border-radius:16px;box-shadow:0 10px 28px rgba(18,38,82,.08);display:block;margin:24px 0 32px;max-width:100%;overflow-x:auto\"><table style=\"border-collapse:collapse;border-spacing:0;font-size:15px;margin:0;width:100%\"><thead><tr><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#08755a;border-color:#287a66;color:#fff;text-align:center\">\u512a\u9ede<\/th><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#b3334b;border-color:#a84b5d;color:#fff;text-align:center\">\u512a\u9ede<\/th><\/tr><\/thead><tbody><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#effaf6;border-left:4px solid #2e9a77;color:#174c3d\">Stable model ID for production use<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#fff3f5;border-left:4px solid #d15d72;color:#632936\">No Live API support<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#effaf6;border-left:4px solid #2e9a77;color:#174c3d\">Lower Standard output price than 3.5 Flash<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#fff3f5;border-left:4px solid #d15d72;color:#632936\">Text output only; no native image or audio generation<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#effaf6;border-left:4px solid #2e9a77;color:#174c3d\">Strong official DeepSWE, MLE-Bench, OSWorld, and long-context gains<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#fff3f5;border-left:4px solid #d15d72;color:#632936\">A 1M-token capacity does not guarantee reliable retrieval<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#effaf6;border-left:4px solid #2e9a77;color:#174c3d\">Broad multimodal and tool support<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#fff3f5;border-left:4px solid #d15d72;color:#632936\">Computer Use remains a Preview capability<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#effaf6;border-left:4px solid #2e9a77;color:#174c3d\">Our bounded extraction, chart, coding, and local action tests passed<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#fff3f5;border-left:4px solid #d15d72;color:#632936\">Our synthetic 128K eight-needle test failed twice<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#effaf6;border-left:4px solid #2e9a77;color:#174c3d\">Batch\/Flex and a cheaper Flash-Lite tier enable routing strategies<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#fff3f5;border-left:4px solid #d15d72;color:#632936\">Reasoning tokens can consume surprisingly tight output caps<\/td><\/tr><\/tbody><\/table><figcaption style=\"background:#f5f7fc;border-top:1px solid #d9e1ef;color:#52617a;font-size:13px;line-height:1.5;margin:0;padding:12px 15px\">Decision summary based on Google documentation, our July 22 test run, and clearly separated early reports.<\/figcaption><\/figure>\n\n\n\n<div class=\"wp-block-group is-layout-constrained wp-block-group-is-layout-constrained\">\n<figure class=\"wp-block-image size-large\"><img alt=\"\" fetchpriority=\"high\" decoding=\"async\" width=\"1024\" height=\"640\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/03\/image-831-1024x640.png\" class=\"wp-image-15877\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/03\/image-831-1024x640.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/03\/image-831-300x187.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/03\/image-831-768x480.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/03\/image-831-1536x960.png 1536w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/03\/image-831-2048x1279.png 2048w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/03\/image-831-18x12.png 18w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<div class=\"wp-block-buttons is-content-justification-center is-layout-flex wp-container-core-buttons-is-layout-3e41869c wp-block-buttons-is-layout-flex\">\n<div class=\"wp-block-button\"><a class=\"wp-block-button__link has-black-color has-luminous-vivid-amber-background-color has-text-color has-background has-link-color wp-element-button\">\u5728 GlobalGPT \u4e0a\u8a66\u7528 100 \u591a\u7a2e\u9802\u7d1a\u6a21\u578b<\/a><\/div>\n<\/div>\n<\/div>\n\n\n\n<h2 id=\"table-of-contents\" class=\"wp-block-heading g36-review\">\u76ee\u9304<\/h2>\n\n\n\n<nav class=\"g36-review g36-toc\" aria-label=\"Article sections\"><ol class=\"g36-toc__grid\"><li><a class=\"g36-toc__link\" href=\"#what-is-gemini-36-flash\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">01<\/span>\u5b83\u662f\u4ec0\u9ebc<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#specifications\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">02<\/span>Specifications<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#pricing\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">03<\/span>\u5b9a\u50f9<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#official-benchmarks\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">04<\/span>Official benchmarks<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#hands-on-tests\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">05<\/span>Hands-on tests<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#api-access\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">06<\/span>API \u5b58\u53d6\u6b0a\u9650<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#migration\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">07<\/span>Migration<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#comparison\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">08<\/span>3.6 vs alternatives<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#early-feedback\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">09<\/span>Early feedback<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#limitations\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">10<\/span>Limits and safety<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#who-should-use\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">11<\/span>Who should use it<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#verdict\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">12<\/span>\u6700\u7d42\u88c1\u6c7a<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#faq\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">13<\/span>\u5e38\u898b\u554f\u984c<\/a><\/li><li><a class=\"g36-toc__link\" href=\"#sources\" style=\"align-items:center;background:#f7f9fd;border:1px solid #d9e1ef;border-radius:14px;box-shadow:0 8px 22px rgba(32,55,105,.07);color:#233558;display:flex;font-weight:750;gap:12px;min-height:62px;padding:13px 16px;text-decoration:none\"><span style=\"align-items:center;background:#e6edfc;border-radius:999px;color:#173e87;display:inline-flex;flex:0 0 34px;font-size:12px;height:34px;justify-content:center\">14<\/span>\u4f86\u6e90<\/a><\/li><\/ol><\/nav>\n\n\n\n<h2 id=\"what-is-gemini-36-flash\" class=\"wp-block-heading g36-review\">What is Gemini 3.6 Flash?<\/h2>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Gemini 3.6 Flash is a stable Google model aimed at production workloads that need more reasoning and tool use than a lightweight inference tier, without moving to a flagship price class. Google describes it as a workhorse for coding, knowledge work, multimodal understanding, and agentic execution. The production model ID is <code>gemini-3.6-flash<\/code>. <a href=\"https:\/\/ai.google.dev\/gemini-api\/docs\/models\/gemini-3.6-flash\">Google&#8217;s official model page<\/a> lists it as generally available.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The model accepts text, images, audio, video, and PDF inputs. It returns text, structured JSON, and function or tool calls, but it does not natively generate images or audio. That distinction matters when comparing broad \u201cmultimodal\u201d claims: understanding multiple input formats is different from producing multiple output media.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Google \u7684 <a href=\"https:\/\/deepmind.google\/models\/model-cards\/gemini-3-6-flash\/\">Gemini 3.6 Flash model card<\/a> gives a March 2026 knowledge cutoff and says the model is based on Gemini 3.5 Flash. Architecture and training details are mostly inherited by reference rather than newly disclosed, so buyers can evaluate documented behavior and published evaluations but cannot reconstruct the training recipe.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The broad tool stack is central to the positioning. The official model page lists function calling, structured outputs, code execution, file search, URL context, context caching, Google Search grounding, Google Maps grounding, thinking, and Computer Use. A production decision should evaluate the whole loop\u2014request shape, tool trace, retries, validation, and human review\u2014not just the quality of one answer.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">That workhorse role also explains the review&#8217;s focus: 3.6 Flash is most interesting when it participates in an application workflow rather than answering an isolated trivia prompt. Its value depends on whether the surrounding system constrains tools, catches truncation, verifies outputs, and routes simpler work to a cheaper tier.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Computer Use lets an application send screenshots and receive proposed actions for browser, mobile, or desktop environments. The client still executes the action, supplies the next screenshot, and enforces policies. Google labels the feature Preview and recommends supervision for sensitive or irreversible work in the <a href=\"https:\/\/ai.google.dev\/gemini-api\/docs\/computer-use\">Computer Use documentation<\/a>.<\/p>\n\n\n\n<h2 id=\"specifications\" class=\"wp-block-heading g36-review\">Gemini 3.6 Flash specifications and capabilities<\/h2>\n\n\n\n<figure class=\"g36-review wp-block-table g36-table g36-table--comparison\" tabindex=\"0\" role=\"region\" aria-label=\"Gemini 3.6 Flash official specifications\" style=\"border:1px solid #d9e1ef;border-radius:16px;box-shadow:0 10px 28px rgba(18,38,82,.08);display:block;margin:24px 0 32px;max-width:100%;overflow-x:auto\"><table style=\"border-collapse:collapse;border-spacing:0;font-size:15px;margin:0;width:100%\"><thead><tr><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom\">\u898f\u683c<\/th><th class=\"g36-highlight\" style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#dceaff;color:#123d85;font-weight:850\">Gemini 3.6 \u9583\u5149\u71c8<\/th><\/tr><\/thead><tbody><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Model ID<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\"><code>gemini-3.6-flash<\/code><\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Release status<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">Stable \/ generally available<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Maximum input context<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">1,048,576 \u500b\u4ee3\u5e63<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Maximum output<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">65,536 \u500b\u4ee3\u5e63<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">\u8f38\u5165\u6a21\u5f0f<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">\u6587\u5b57\u3001\u5f71\u50cf\u3001\u97f3\u8a0a\u3001\u8996\u8a0a\u3001PDF<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Output modality<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">\u6587\u672c<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Default thinking level<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\"><code>\u4e2d\u578b<\/code><\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">\u77e5\u8b58\u622a\u6b62<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">March 2026<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Live API<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">\u4e0d\u652f\u63f4<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">\u96fb\u8166\u4f7f\u7528<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">Supported in Preview<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">\u6279\u6b21 API<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">Supported at model level; not yet through Interactions API<\/td><\/tr><\/tbody><\/table><figcaption style=\"background:#f5f7fc;border-top:1px solid #d9e1ef;color:#52617a;font-size:13px;line-height:1.5;margin:0;padding:12px 15px\">Google official model specifications, accessed July 22, 2026.<\/figcaption><\/figure>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The 1,048,576-token context limit is capacity, not a promise that every fact can be recovered reliably. This is a recurring theme in both Google&#8217;s long-context benchmarks and our synthetic gateway test. Retrieval, chunking, document indexes, summaries, and source citations remain useful even when an entire corpus technically fits in one request.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The 65,536-token output ceiling is generous, but real applications often set a lower cap for cost and latency. Our tests exposed how a cap that looks adequate for the visible answer can be consumed by internal reasoning tokens. Validate the actual finish reason and content, rather than treating an HTTP 200 response as task success.<\/p>\n\n\n\n<h2 id=\"pricing\" class=\"wp-block-heading g36-review\">Gemini 3.6 Flash pricing<\/h2>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Official paid Standard pricing is <strong>$1.50 per million input tokens<\/strong> \u548c <strong>$7.50 per million output tokens<\/strong>. Output billing includes thinking tokens. Batch and Flex inference halve those token rates, while Priority inference applies an 80% premium. Always verify the current <a href=\"https:\/\/ai.google.dev\/gemini-api\/docs\/pricing\">Gemini API \u5b9a\u50f9\u9801\u9762<\/a> before forecasting a production bill.<\/p>\n\n\n<p><!-- INTERNAL LINK TODO: add the site's Gemini API pricing comparison after this paragraph when the canonical URL is known. --><\/p>\n\n\n<figure class=\"g36-review wp-block-table g36-table g36-table--pricing\" tabindex=\"0\" role=\"region\" aria-label=\"Gemini 3.6 Flash API pricing tiers\" style=\"border:1px solid #d9e1ef;border-radius:16px;box-shadow:0 10px 28px rgba(18,38,82,.08);display:block;margin:24px 0 32px;max-width:100%;overflow-x:auto\"><table style=\"border-collapse:collapse;border-spacing:0;font-size:15px;margin:0;width:100%\"><thead><tr><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#6040a3;border-color:#765eb0;color:#fff;text-align:center\">Consumption mode<\/th><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#6040a3;border-color:#765eb0;color:#fff;text-align:center\">\u8f38\u5165 \/ 100 \u842c\u500b\u4ee3\u5e63<\/th><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#6040a3;border-color:#765eb0;color:#fff;text-align:center\">\u7522\u51fa \/ 100\u842c\u500b\u4ee3\u5e63<\/th><\/tr><\/thead><tbody><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#eef4ff;font-variant-numeric:tabular-nums\">\u6a19\u6e96<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#eef4ff;font-variant-numeric:tabular-nums\">$1.50<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#eef4ff;font-variant-numeric:tabular-nums\">$7.50<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#eaf8f4;font-variant-numeric:tabular-nums\">\u6279\u6b21<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#eaf8f4;font-variant-numeric:tabular-nums\">$0.75<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#eaf8f4;font-variant-numeric:tabular-nums\">$3.75<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#f3efff;font-variant-numeric:tabular-nums\">Flex<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#f3efff;font-variant-numeric:tabular-nums\">$0.75<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#f3efff;font-variant-numeric:tabular-nums\">$3.75<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#fff4df;font-variant-numeric:tabular-nums\">\u512a\u5148\u9806\u5e8f<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#fff4df;font-variant-numeric:tabular-nums\">$2.70<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#fff4df;font-variant-numeric:tabular-nums\">$13.50<\/td><\/tr><\/tbody><\/table><figcaption style=\"background:#f5f7fc;border-top:1px solid #d9e1ef;color:#52617a;font-size:13px;line-height:1.5;margin:0;padding:12px 15px\">USD list prices reported by Google on July 22, 2026; service availability and limits may vary.<\/figcaption><\/figure>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Context caching is priced at $0.15 per million cached input tokens in the Standard paid tier, plus $1.00 per million tokens per hour for cache storage. Caching can reduce repeated-input charges, but a large cache retained longer than needed becomes its own cost center. Compare the cache bill with the reuse actually achieved.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Grounding has a different unit. Google lists 5,000 Google Search or Maps grounded prompts per month at no charge, shared across Gemini 3 models, followed by $14 per 1,000 search queries. One prompt may issue more than one query, so token cost alone can understate a grounded workflow.<\/p>\n\n\n\n<h3 id=\"pricing-example\" class=\"wp-block-heading g36-review\">A $2.25 Standard pricing example<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">A request that consumes 1,000,000 input tokens and produces 100,000 output tokens costs $1.50 for input plus $0.75 for output, or <strong>$2.25<\/strong> in Standard mode. That simple example excludes cache storage, grounding queries, retries, external tools, and failed attempts.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Google also reports that Gemini 3.6 Flash used 17% fewer output tokens than 3.5 Flash in the Artificial Analysis Index and needed fewer reasoning steps and tool calls in multi-step workflows. That is an official workload-dependent observation, not a guaranteed 17% saving for every application. Measure tokens per successful task with your own prompts.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" width=\"1024\" height=\"627\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/pricing-card-1-1024x627.png\" alt=\"Official list-price summary with the $2.25 Standard example. Confirm current prices before publishing or budgeting.\" class=\"wp-image-16931\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/pricing-card-1-1024x627.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/pricing-card-1-300x184.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/pricing-card-1-768x470.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/pricing-card-1-1536x941.png 1536w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/pricing-card-1-18x12.png 18w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/pricing-card-1.png 1600w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n<p><!-- \u4e0a\u4f20\u672c\u5730\u6587\u4ef6 output\/gemini-3-6-flash-wp\/visuals\/pricing-card.png\uff1b\u5982\u5a92\u4f53\u5e93\u751f\u6210\u7684 URL \u4e0d\u540c\uff0c\u8bf7\u66ff\u6362\u4e0b\u65b9 src\u3002 --><\/p>\n\n\n<h2 id=\"official-benchmarks\" class=\"wp-block-heading g36-review\">Gemini 3.6 Flash benchmarks: what Google reports<\/h2>\n\n\n\n<div class=\"g36-review g36-evidence-key\" aria-label=\"Evidence labels\" style=\"display:flex;flex-wrap:wrap;gap:9px;margin:18px 0 24px\"><span class=\"g36-badge g36-badge--official\" style=\"border:1px solid transparent;border-radius:999px;display:inline-flex;font-size:12px;font-weight:800;padding:6px 11px;background:#e2edff;border-color:#bed2f7;color:#194584\">Google official result<\/span><span class=\"g36-badge g36-badge--ours\" style=\"border:1px solid transparent;border-radius:999px;display:inline-flex;font-size:12px;font-weight:800;padding:6px 11px;background:#e2f5ed;border-color:#b8dfcf;color:#135d44\">Our Broly gateway test<\/span><span class=\"g36-badge g36-badge--third\" style=\"border:1px solid transparent;border-radius:999px;display:inline-flex;font-size:12px;font-weight:800;padding:6px 11px;background:#f0eafd;border-color:#d5c5f2;color:#553493\">Third-party benchmark<\/span><span class=\"g36-badge g36-badge--social\" style=\"border:1px solid transparent;border-radius:999px;display:inline-flex;font-size:12px;font-weight:800;padding:6px 11px;background:#fff0da;border-color:#efd1a4;color:#704300\">Social anecdote<\/span><\/div>\n\n\n<p><!-- Google \u7039\u6a3b\u67df\u6fb6\u682d\u647c\u9365\u5267\u5896\u951b\u5c7e\u68e4\u95c7\u20ac\u6d93\u5a41\u7d36\u951b\u6cb4rc=https:\/\/storage.googleapis.com\/gweb-uniblog-publish-prod\/images\/gemini-3-6-flash__evals__figure-.width-1200.format-webp.webp --><\/p>\n\n\n<figure class=\"wp-block-image size-full g36-review g36-figure g36-figure--official\"><a href=\"https:\/\/blog.google\/innovation-and-ai\/models-and-research\/gemini-models\/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber\/\"><img decoding=\"async\" src=\"https:\/\/storage.googleapis.com\/gweb-uniblog-publish-prod\/images\/gemini-3-6-flash__evals__figure-.width-1200.format-webp.webp\" alt=\"Google official Gemini 3.6 Flash evaluation chart published with the model release\"\/><\/a><figcaption class=\"wp-element-caption\">Official Gemini 3.6 Flash evaluation figure published by Google; benchmark names and methodology remain attached to the source. <a href=\"https:\/\/blog.google\/innovation-and-ai\/models-and-research\/gemini-models\/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber\/\">View the official Google source<\/a>.<\/figcaption><\/figure>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Google&#8217;s official comparisons show the clearest gains in agentic coding, machine-learning engineering, computer interaction, and long-context retrieval. The most useful baseline is Gemini 3.5 Flash because Google evaluated both models within the same model-card family. These values are Google&#8217;s results, not measurements from our test harness.<\/p>\n\n\n\n<figure class=\"g36-review wp-block-table g36-table g36-table--comparison\" tabindex=\"0\" role=\"region\" aria-label=\"Google official benchmark comparison\" style=\"border:1px solid #d9e1ef;border-radius:16px;box-shadow:0 10px 28px rgba(18,38,82,.08);display:block;margin:24px 0 32px;max-width:100%;overflow-x:auto\"><table style=\"border-collapse:collapse;border-spacing:0;font-size:15px;margin:0;width:100%\"><thead><tr><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom\">\u57fa\u6e96<\/th><th class=\"g36-highlight\" style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#dceaff;color:#123d85;font-weight:850\">Gemini 3.6 \u9583\u5149\u71c8<\/th><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom\">\u96d9\u5b50\u661f 3.5 Flash<\/th><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom\">Reported difference<\/th><\/tr><\/thead><tbody><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">SWE-Bench Pro Public<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">58.7%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">55.1%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">+3.6 points<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">DeepSWE v1.1<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">49.0%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">37.0%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">+12.0 points<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Terminal-Bench 2.1<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">78.0%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">76.2%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">+1.8 points<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">MLE-Bench<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">63.9%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">49.7%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">+14.2 points<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">GDPval-AA v2<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">1,421 Elo<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">1,349 Elo<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">+72 Elo<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">OSWorld-Verified<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">83.0%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">78.4%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">+4.6 points<\/td><\/tr><\/tbody><\/table><figcaption style=\"background:#f5f7fc;border-top:1px solid #d9e1ef;color:#52617a;font-size:13px;line-height:1.5;margin:0;padding:12px 15px\">Google official\/model-card values. Different benchmarks measure different task distributions and are not interchangeable.<\/figcaption><\/figure>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">DeepSWE improved by 12.0 percentage points and MLE-Bench by 14.2 points. The OSWorld-Verified gain was 4.6 points, while SWE-Bench Pro Public and Terminal-Bench moved more modestly. The defensible claim is that 3.6 improves several agentic workloads relative to 3.5, not that it dominates every model or coding benchmark.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Google also describes fewer unwanted code edits and shorter execution loops. Those behaviors can matter more than a small pass-rate change in repository work because unnecessary changes add review cost and repeated loops consume tokens. They still require application-level measurement; an aggregate benchmark cannot establish behavior in a specific codebase.<\/p>\n\n\n\n<h3 id=\"multimodal-benchmarks\" class=\"wp-block-heading g36-review\">Multimodal and long-context results<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">On CharXIV reasoning over complex charts, Google reports 85.2% without tools and 89.4% with tools for Gemini 3.6 Flash. Gemini 3.5 Flash scored 84.2% and 84.9% respectively. The larger tool-assisted delta supports the model&#8217;s positioning for chart interpretation and multimodal workflows, while not measuring visual taste or front-end styling quality.<\/p>\n\n\n\n<figure class=\"g36-review wp-block-table g36-table g36-table--comparison\" tabindex=\"0\" role=\"region\" aria-label=\"Google long-context benchmark comparison\" style=\"border:1px solid #d9e1ef;border-radius:16px;box-shadow:0 10px 28px rgba(18,38,82,.08);display:block;margin:24px 0 32px;max-width:100%;overflow-x:auto\"><table style=\"border-collapse:collapse;border-spacing:0;font-size:15px;margin:0;width:100%\"><thead><tr><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom\">GDM-MRCR v2 long-context test<\/th><th class=\"g36-highlight\" style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#dceaff;color:#123d85;font-weight:850\">Gemini 3.6 \u9583\u5149\u71c8<\/th><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom\">\u96d9\u5b50\u661f 3.5 Flash<\/th><\/tr><\/thead><tbody><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">128K, 8-needle average<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">91.8%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">77.3%<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">1M, 8-needle pointwise<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">54.0%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">26.6%<\/td><\/tr><\/tbody><\/table><figcaption style=\"background:#f5f7fc;border-top:1px solid #d9e1ef;color:#52617a;font-size:13px;line-height:1.5;margin:0;padding:12px 15px\">Google official GDM-MRCR v2 figures. Context capacity and retrieval reliability are different properties.<\/figcaption><\/figure>\n\n\n<p><!-- Google \u7039\u6a3b\u67df\u6fb6\u682d\u647c\u9365\u5267\u5896\u951b\u5c7e\u68e4\u95c7\u20ac\u6d93\u5a41\u7d36\u951b\u6cb4rc=https:\/\/storage.googleapis.com\/gweb-uniblog-publish-prod\/images\/gemini-3-6-flash__evals__quality.width-1200.format-webp.webp --><\/p>\n\n\n<figure class=\"wp-block-image size-full g36-review g36-figure g36-figure--official\"><a href=\"https:\/\/blog.google\/innovation-and-ai\/models-and-research\/gemini-models\/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber\/\"><img decoding=\"async\" src=\"https:\/\/storage.googleapis.com\/gweb-uniblog-publish-prod\/images\/gemini-3-6-flash__evals__quality.width-1200.format-webp.webp\" alt=\"Google official Gemini 3.6 Flash quality evaluation chart from the release article\"\/><\/a><figcaption class=\"wp-element-caption\">Official Gemini 3.6 Flash quality figure published by Google; interpret it alongside the release article and model card. <a href=\"https:\/\/blog.google\/innovation-and-ai\/models-and-research\/gemini-models\/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber\/\">View the official Google source<\/a>.<\/figcaption><\/figure>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The 1M result is a large relative improvement, but 54.0% is not a basis for blind retrieval of critical facts. Production systems should still segment documents, retain provenance, and ask for source citations. Our separate 128K synthetic test failed, but it used a different task and gateway route and cannot be substituted for Google&#8217;s result.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" width=\"1024\" height=\"730\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/benchmark-deltas-1024x730.png\" alt=\"Google-reported deltas for DeepSWE, MLE-Bench, OSWorld-Verified, and 1M long-context retrieval; these are not our test results.\" class=\"wp-image-16932\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/benchmark-deltas-1024x730.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/benchmark-deltas-300x214.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/benchmark-deltas-768x547.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/benchmark-deltas-1536x1094.png 1536w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/benchmark-deltas-18x12.png 18w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/benchmark-deltas.png 1600w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n<p><!-- INTERNAL LINK TODO: add the site's coding-model benchmark guide here when its canonical URL is available. --><\/p>\n<p><!-- \u4e0a\u4f20\u672c\u5730\u6587\u4ef6 output\/gemini-3-6-flash-wp\/visuals\/benchmark-deltas.png\uff1b\u5982\u5a92\u4f53\u5e93\u751f\u6210\u7684 URL \u4e0d\u540c\uff0c\u8bf7\u66ff\u6362\u4e0b\u65b9 src\u3002 --><\/p>\n\n\n<h2 id=\"hands-on-tests\" class=\"wp-block-heading g36-review\">Our hands-on Gemini 3.6 Flash tests<\/h2>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">We ran a deterministic first-party suite on July 22, 2026. Every model call went through <code>https:\/\/anywhere.broly.ai\/v1<\/code>; we called no native Google endpoint. Search Grounding was disabled. All reported latency is <strong>gateway-inclusive total response time<\/strong>, so it includes Broly routing overhead and must not be read as direct Google API latency.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">TTFT was not measured, including for the streaming check. We recorded total elapsed time, prompt tokens, completion tokens, reasoning tokens, cached tokens, total tokens, finish reason, retries, validation, and an estimated cost based on the official Standard token rates. Reasoning tokens were already included in completion tokens, and cached tokens were already included in prompt tokens.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The complete activity comprised <strong>12 logical API calls<\/strong> \u548c <strong>14 HTTP attempts<\/strong>. The primary run remained <strong>$0.244491<\/strong>, the three targeted reruns totaled <strong>$0.0520416<\/strong>, and the corrected cumulative estimate was <strong>$0.2965326<\/strong>. After replacing two clearly truncated initial records with their targeted reruns, the consolidated outcome was: <strong>7 passed, 1 failed, and 1 was unsupported<\/strong>.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Cost estimates apply $1.50\/M for uncached prompt tokens, $0.15\/M for cached prompt tokens, and $7.50\/M for output tokens. In the long-context rerun, 128,248 prompt tokens split into 5,464 uncached and 122,784 cached tokens; adding 2,048 output tokens produces <strong>$0.0419736<\/strong>. Cache storage is listed at $1.00\/M tokens\/hour but is excluded because storage duration was not measured.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">These are small-sample diagnostic tests, not a statistically powered benchmark. They are useful because the fixtures, expected values, validators, request limits, and failures are preserved. They should guide follow-up testing in a real workload, not become a universal quality score.<\/p>\n\n\n\n<figure class=\"g36-review wp-block-table g36-table g36-table--results\" tabindex=\"0\" role=\"region\" aria-label=\"Broly gateway hands-on test results\" style=\"border:1px solid #d9e1ef;border-radius:16px;box-shadow:0 10px 28px rgba(18,38,82,.08);display:block;margin:24px 0 32px;max-width:100%;overflow-x:auto\"><table style=\"border-collapse:collapse;border-spacing:0;font-size:15px;margin:0;width:100%\"><thead><tr><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#087f78;border-color:#318d86;color:#fff\">Final task record<\/th><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#087f78;border-color:#318d86;color:#fff\">\u7d50\u679c<\/th><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#087f78;border-color:#318d86;color:#fff\">Gateway-inclusive time<\/th><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#087f78;border-color:#318d86;color:#fff\">Tokens P \/ C \/ R \/ Cached \/ Total<\/th><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#087f78;border-color:#318d86;color:#fff\">\u4f30\u8a08\u6210\u672c<\/th><\/tr><\/thead><tbody><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Streaming exact answer, 3.6<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\"><span class=\"g36-status g36-status--pass\" style=\"background:#dff5eb;border-radius:999px;color:#0d5b40;display:inline-block;font-size:11px;font-weight:850;padding:4px 9px;text-transform:uppercase\">Pass<\/span><\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">2.938 seconds<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">5 \/ 119 \/ 118 \/ 0 \/ 124<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">$0.0009<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Structured extraction, 3.6 rerun<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\"><span class=\"g36-status g36-status--pass\" style=\"background:#dff5eb;border-radius:999px;color:#0d5b40;display:inline-block;font-size:11px;font-weight:850;padding:4px 9px;text-transform:uppercase\">Pass<\/span><\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">3.919 seconds<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">55 \/ 453 \/ 411 \/ 0 \/ 508<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">$0.00348<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Structured extraction, 3.5<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\"><span class=\"g36-status g36-status--pass\" style=\"background:#dff5eb;border-radius:999px;color:#0d5b40;display:inline-block;font-size:11px;font-weight:850;padding:4px 9px;text-transform:uppercase\">Pass<\/span><\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">3.113 seconds<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">56 \/ 378 \/ 324 \/ 0 \/ 434<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">$0.003486<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Chart vision, 3.6 rerun<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\"><span class=\"g36-status g36-status--pass\" style=\"background:#dff5eb;border-radius:999px;color:#0d5b40;display:inline-block;font-size:11px;font-weight:850;padding:4px 9px;text-transform:uppercase\">Pass<\/span><\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">4.884 seconds<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">1,127 \/ 653 \/ 520 \/ 0 \/ 1,780<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">$0.006588<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Coding repair, 3.6<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\"><span class=\"g36-status g36-status--pass\" style=\"background:#dff5eb;border-radius:999px;color:#0d5b40;display:inline-block;font-size:11px;font-weight:850;padding:4px 9px;text-transform:uppercase\">Pass<\/span><\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">5.553 seconds<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">602 \/ 1,150 \/ 930 \/ 0 \/ 1,752<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">$0.009528<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Coding repair, 3.5<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\"><span class=\"g36-status g36-status--pass\" style=\"background:#dff5eb;border-radius:999px;color:#0d5b40;display:inline-block;font-size:11px;font-weight:850;padding:4px 9px;text-transform:uppercase\">Pass<\/span><\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">6.860 seconds<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">603 \/ 1,332 \/ 1,058 \/ 0 \/ 1,935<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">$0.0128925<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">128K eight-needle, 3.6 rerun<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\"><span class=\"g36-status g36-status--fail\" style=\"background:#fee9ed;border-radius:999px;color:#8a2439;display:inline-block;font-size:11px;font-weight:850;padding:4px 9px;text-transform:uppercase\">Fail<\/span><\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">11.366 seconds<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">128,248 \/ 2,048 \/ 0 \/ 122,784 \/ 130,296<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">$0.0419736<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Local browser action plan, 3.6<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\"><span class=\"g36-status g36-status--pass\" style=\"background:#dff5eb;border-radius:999px;color:#0d5b40;display:inline-block;font-size:11px;font-weight:850;padding:4px 9px;text-transform:uppercase\">Pass<\/span><\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">5.305 seconds<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">1,153 \/ 643 \/ 573 \/ 0 \/ 1,796<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">$0.006552<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Broly <code>\/responses<\/code> PDF conversion<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\"><span class=\"g36-status g36-status--unsupported\" style=\"background:#fff0cf;border-radius:999px;color:#774900;display:inline-block;font-size:11px;font-weight:850;padding:4px 9px;text-transform:uppercase\">Unsupported<\/span><\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">2.423 seconds<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">0 \/ 0 \/ 0 \/ 0 \/ 0<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">$0<\/td><\/tr><\/tbody><\/table><figcaption style=\"background:#f5f7fc;border-top:1px solid #d9e1ef;color:#52617a;font-size:13px;line-height:1.5;margin:0;padding:12px 15px\">Our Broly gateway test records. P = prompt, C = completion, R = reasoning; reasoning is included in completion. Cached prompt tokens use the official cached-input rate.<\/figcaption><\/figure>\n\n\n<p><!-- \u4e0a\u4f20\u672c\u5730\u6587\u4ef6 output\/gemini-3-6-flash-wp\/hands-on\/screenshots\/test-dashboard-summary.png\uff1b\u5982\u5a92\u4f53\u5e93\u751f\u6210\u7684 URL \u4e0d\u540c\uff0c\u8bf7\u66ff\u6362\u4e0b\u65b9 src\u3002 --><\/p>\n\n<p><!-- \u8bf7\u4e0a\u4f20\u672c\u5730\u6587\u4ef6 output\/gemini-3-6-flash-wp\/hands-on\/screenshots\/test-dashboard.png\uff0c\u5e76\u4fdd\u6301\u539f\u59cb 1600\u00d73509 \u5c3a\u5bf8\uff1b\u5982\u5a92\u4f53\u5e93 URL \u4e0d\u540c\uff0c\u8bf7\u66ff\u6362\u4e0b\u65b9 href\uff0c\u4ec5\u7528\u4e8e\u653e\u5927\u67e5\u770b\u6216\u4e0b\u8f7d\uff0c\u4e0d\u8981\u7a84\u680f\u5d4c\u5165\u3002 --><\/p>\n\n\n<h3 id=\"streaming-structured-chart\" class=\"wp-block-heading g36-review\">Streaming, structured output, and chart reading<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The streaming task asked for the exact visible answer <code>OK<\/code>. Gemini 3.6 Flash returned the correct answer in 2.938 seconds with 5 prompt, 119 completion, 118 reasoning, and 124 total tokens, at an estimated $0.0009. The striking detail is that reasoning consumed 118 of 119 completion tokens, illustrating why a tiny output cap can leave no visible answer.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The first 3.6 structured-extraction call used a 512-token output cap and ended at <code>finish_reason=length<\/code> with truncated JSON. We treated that as a harness limit, not a quality verdict, and repeated only that task at 2,048 tokens. The targeted rerun passed exact field validation in 3.919 seconds with P55\/C453\/R411\/T508 and an estimated $0.00348.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Gemini 3.5 Flash passed the same extraction on its initial call in 3.113 seconds with P56\/C378\/R324\/T434 and an estimated $0.003486. One run per configuration cannot establish a latency winner. It does show that output limits and reasoning usage should be tuned per model instead of copied blindly.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The 3.6 chart-reading call was also truncated at an initial cap of 768 tokens. At 2,048 tokens, the targeted rerun read all eight benchmark values exactly in 4.884 seconds, using P1,127\/C653\/R520\/T1,780 and an estimated $0.006588. The validator compared every returned value with a fixed ground-truth file.<\/p>\n\n\n\n<h3 id=\"coding-repair\" class=\"wp-block-heading g36-review\">Coding repair: both models reached 5\/5<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Both coding tasks started from isolated copies of the same Python fixture with a baseline of 3\/5 passing tests. Gemini 3.6 Flash produced a focused usable repair in 5.553 seconds, with P602\/C1,150\/R930\/T1,752 and an estimated $0.009528. Applying the output raised the fixture to 5\/5, while the test file hash remained unchanged.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Gemini 3.5 Flash also produced a focused repair that reached 5\/5, in 6.860 seconds with P603\/C1,332\/R1,058\/T1,935 and an estimated $0.0128925. It added prose around fenced code despite the no-prose instruction, so the harness needed tolerant fenced-code extraction before offline validation. Both repairs were usable; output discipline differed.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">This test favors bounded repair work with deterministic unit tests. It does not measure repository navigation, multi-file architectural changes, long-running autonomy, or whether the model can design a discriminating test from scratch. Those dimensions need separate fixtures and failure criteria.<\/p>\n\n\n\n<h3 id=\"long-context-test\" class=\"wp-block-heading g36-review\">The 128K synthetic retrieval failure<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Our near-128K prompt placed eight deterministic identifiers in a large synthetic context and required exact JSON recovery. Gemini 3.6 Flash failed twice, returning unrelated code and HTML rather than the eight values. Raising the output cap did not correct the behavior.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The rerun took 11.366 seconds and reported P128,248\/C2,048\/R0\/Cached122,784\/T130,296, an estimated $0.0419736 after the cached-input adjustment, and <code>finish_reason=length<\/code>. This is <strong>not directly comparable to Google\u2019s GDM-MRCR<\/strong> result: our task, exact prompt construction, context scale, output format, and Broly gateway route all differ.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The failure is still operationally relevant. It says this particular synthetic request did not work through our route on two attempts, so we would not ship the same pattern without retrieval, tighter context selection, response validation, and fallback handling. It does not erase Google&#8217;s benchmark or prove a general 128K limitation.<\/p>\n\n\n\n<h3 id=\"browser-harness\" class=\"wp-block-heading g36-review\">Local browser-action harness: correct target, invented ID<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The model inspected a screenshot of a controlled local page and correctly identified the visible target text, \u201cOpen usage report.\u201d It proposed the action <code>click<\/code> but invented a hidden identifier, <code>open_usage_report<\/code>, instead of the real DOM ID <code>\u7528\u6cd5<\/code>. The harness resolved the unique exact visible text rather than trusting the invented ID.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The plan call took 5.305 seconds with P1,153\/C643\/R573\/T1,796 and cost an estimated $0.006552. A local Chrome fixture was clicked once, changing state from <code>ready<\/code> \u81f3 <code>report-open<\/code>. This was <strong>not native Gemini Computer Use<\/strong>; it made no external action, submitted no form, and changed no external state.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"720\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/browser-before-1024x720.png\" alt=\"Before: isolated local fixture in the ready state; no external page or form was involved.\" class=\"wp-image-16934\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/browser-before-1024x720.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/browser-before-300x211.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/browser-before-768x540.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/browser-before-18x12.png 18w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/browser-before.png 1280w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"653\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/browser-after-1024x653.png\" alt=\"After: the exact visible-text resolver clicked local ID usage, producing the verified report-open state.\" class=\"wp-image-16937\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/browser-after-1024x653.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/browser-after-300x191.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/browser-after-768x490.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/browser-after-18x12.png 18w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/browser-after.png 1256w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n<p><!-- \u4e0a\u4f20\u672c\u5730\u6587\u4ef6 output\/gemini-3-6-flash-wp\/hands-on\/screenshots\/browser-before.png\uff1b\u5982\u5a92\u4f53\u5e93\u751f\u6210\u7684 URL \u4e0d\u540c\uff0c\u8bf7\u66ff\u6362\u4e0b\u65b9 src\u3002 --><\/p>\n\n<p><!-- \u4e0a\u4f20\u672c\u5730\u6587\u4ef6 output\/gemini-3-6-flash-wp\/hands-on\/screenshots\/browser-after.png\uff1b\u5982\u5a92\u4f53\u5e93\u751f\u6210\u7684 URL \u4e0d\u540c\uff0c\u8bf7\u66ff\u6362\u4e0b\u65b9 src\u3002 --><\/p>\n\n\n<h3 id=\"pdf-path\" class=\"wp-block-heading g36-review\">The PDF result is a gateway-path limitation<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The Broly <code>\/responses<\/code> PDF conversion path returned HTTP 500 with code <code>convert_request_failed<\/code> and message <code>not implemented<\/code>. The logical call made three HTTP attempts because the record includes two retries, then stopped without token usage or cost. We classify this one route as unsupported.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">This result <strong>does not show that Gemini lacks PDF support<\/strong>. Google&#8217;s model documentation lists PDF input; our request failed before a visible model response on an OpenAI-compatible gateway conversion path. A native Google AI Studio or Gemini API check is still required before publishing any conclusion about native PDF behavior.<\/p>\n\n\n<p><!-- \u7f16\u8f91\u64cd\u4f5c\uff1a\u4ecd\u9700\u5728 Google AI Studio \u6216\u539f\u751f Gemini API \u8865\u6d4b PDF\u3002\u8bf7\u4fdd\u5b58\u56db\u5f20\u539f\u751f\u8bc1\u636e\u622a\u56fe\uff1a\u5b8c\u6574\u63d0\u793a\u8bcd\u3001\u6a21\u578b\u56de\u7b54\u3001usage\/\u4ee4\u724c\u9762\u677f\u3001\u5e26\u9875\u7801\u5f15\u7528\u7684 PDF \u9875\u9762\uff1b\u53d1\u5e03\u524d\u5c06\u7ed3\u679c\u4e0e\u672c\u6bb5\u6838\u5bf9\uff0c\u7edd\u4e0d\u80fd\u628a Broly \u8f6c\u6362\u5931\u8d25\u5199\u6210 Gemini \u4e0d\u652f\u6301 PDF\u3002 --><\/p>\n\n\n<p class=\"g36-review g36-operation-note wp-block-paragraph\"><strong>Editor evidence still needed:<\/strong> run the PDF task through native Google AI Studio or the Gemini API, then capture the full prompt, model response, usage panel, and page-citation evidence. Until those screenshots exist, this review reports only the Broly conversion failure and makes no native-PDF performance claim.<\/p>\n\n\n\n<h2 id=\"api-access\" class=\"wp-block-heading g36-review\">Gemini 3.6 Flash API access and code<\/h2>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Gemini 3.6 Flash is available in Google AI Studio and the Gemini Developer API under the stable model ID <code>gemini-3.6-flash<\/code>. Google recommends the newer <a href=\"https:\/\/ai.google.dev\/gemini-api\/docs\/interactions\">Interactions API<\/a> for new agentic projects, while the established <code>generateContent<\/code> interface remains supported and covers important features not yet exposed through Interactions.<\/p>\n\n\n<p><!-- INTERNAL LINK TODO: add the site's Google AI Studio setup guide here after its final URL is known. --><\/p>\n\n\n<h3 id=\"python-example\" class=\"wp-block-heading g36-review\">Python with the official Google Gen AI SDK<\/h3>\n\n\n\n<div class=\"g36-review g36-code-panel\" style=\"background:#0d1730;border:1px solid #314b79;border-radius:16px;box-shadow:0 18px 42px rgba(9,20,47,.20);margin:20px 0 28px;overflow:hidden\"><div style=\"align-items:center;background:#10213d;border-bottom:1px solid #26375d;color:#a9bad4;display:flex;font-size:12px;font-weight:800;gap:8px;letter-spacing:.06em;padding:12px 16px\"><span style=\"background:#ff6b6b;border-radius:50%;height:12px;width:12px\"><\/span><span style=\"background:#ffd166;border-radius:50%;height:12px;width:12px\"><\/span><span style=\"background:#45d6ad;border-radius:50%;height:12px;width:12px\"><\/span><span style=\"margin-left:8px\">API example<\/span><\/div><pre style=\"background:#0d1730;color:#eef3ff;font-family:SFMono-Regular,Consolas,Liberation Mono,monospace;font-size:14px;line-height:1.65;margin:0;overflow-x:auto;padding:22px;white-space:pre\"><code style=\"background:transparent;border:0;color:inherit;padding:0\">from google import genai\n\nclient = genai.Client()\n\nresponse = client.models.generate_content(\n    model=\"gemini-3.6-flash\",\n    contents=(\n        \"Review this proposed code change. Identify the root cause, \"\n        \"the smallest safe repair, and the exact test that would prove it.\"\n    ),\n)\n\nprint(response.text)<\/code><\/pre><\/div>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The client reads credentials from the supported environment configuration; do not hard-code or publish a key. The example intentionally avoids deprecated sampling fields and uses Google&#8217;s official <code>google-genai<\/code> <code>generate_content<\/code> method. Add schemas, safety settings, and tools only as the workload requires.<\/p>\n\n\n\n<h3 id=\"javascript-example\" class=\"wp-block-heading g36-review\">JavaScript with <code>@google\/genai<\/code><\/h3>\n\n\n\n<div class=\"g36-review g36-code-panel\" style=\"background:#0d1730;border:1px solid #314b79;border-radius:16px;box-shadow:0 18px 42px rgba(9,20,47,.20);margin:20px 0 28px;overflow:hidden\"><div style=\"align-items:center;background:#10213d;border-bottom:1px solid #26375d;color:#a9bad4;display:flex;font-size:12px;font-weight:800;gap:8px;letter-spacing:.06em;padding:12px 16px\"><span style=\"background:#ff6b6b;border-radius:50%;height:12px;width:12px\"><\/span><span style=\"background:#ffd166;border-radius:50%;height:12px;width:12px\"><\/span><span style=\"background:#45d6ad;border-radius:50%;height:12px;width:12px\"><\/span><span style=\"margin-left:8px\">API example<\/span><\/div><pre style=\"background:#0d1730;color:#eef3ff;font-family:SFMono-Regular,Consolas,Liberation Mono,monospace;font-size:14px;line-height:1.65;margin:0;overflow-x:auto;padding:22px;white-space:pre\"><code style=\"background:transparent;border:0;color:inherit;padding:0\">import { GoogleGenAI } from \"@google\/genai\";\n\nconst ai = new GoogleGenAI({});\n\nconst response = await ai.models.generateContent({\n  model: \"gemini-3.6-flash\",\n  contents: \"Extract the risks and return a concise, source-grounded summary.\",\n});\n\nconsole.log(response.text);<\/code><\/pre><\/div>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">For reliable production use, validate the returned structure, inspect finish reasons, record tool calls, and distinguish transport success from task success. A 200 response containing truncated JSON or an unsupported action is not a passing business outcome.<\/p>\n\n\n\n<h3 id=\"api-choice\" class=\"wp-block-heading g36-review\">Interactions API or generateContent?<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The Interactions API adds optional server-side state through <code>previous_interaction_id<\/code>, observable execution steps, and background execution for longer tasks. It is attractive for new agent loops because state and tool traces are first-class concepts. Review storage and retention settings before using it with sensitive content.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">\u4fdd\u6301 <code>generateContent<\/code> when its mature feature surface better fits the application. As of July 2026, Google&#8217;s documentation says Interactions does not yet expose Batch API, explicit caching, custom safety settings, video metadata, or automatic Python function calling. Model support and API-surface support are separate questions.<\/p>\n\n\n\n<h2 id=\"migration\" class=\"wp-block-heading g36-review\">Gemini 3.6 Flash API migration changes<\/h2>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Moving from earlier request shapes is more than swapping the model name. Google&#8217;s <a href=\"https:\/\/ai.google.dev\/gemini-api\/docs\/latest-model\">latest-model migration guide<\/a> documents parameters and turn patterns that are deprecated, ignored, or rejected. Treat the migration as a request-contract and state-integrity change.<\/p>\n\n\n\n<h3 id=\"migration-parameters\" class=\"wp-block-heading g36-review\">Remove deprecated controls and prefilling<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">\u79fb\u9664 <code>\u6eab\u5ea6<\/code>, <code>top_p<\/code>, \u4ee5\u53ca <code>top_k<\/code>. Google says they are deprecated and ignored for these models, with future generations expected to return HTTP 400 when they are supplied. A legacy <code>temperature: 0<\/code> field should not be treated as a determinism guarantee.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Requests also may not end with a non-empty model-role prefill. Patterns such as appending \u201cTranslation:\u201d or the opening brace of JSON to a model turn can return HTTP 400. Replace prefilling with system instructions, structured outputs, explicit schemas, and validators.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Replace numeric <code>thinking_budget<\/code> controls with the <code>\u601d\u8003\u5c64\u7d1a<\/code> enum; Gemini 3.6 Flash defaults to <code>\u4e2d\u578b<\/code>. Remove <code>candidate_count<\/code>, which Gemini 3.x does not support, and implement application-level alternatives if multiple candidate strategies are genuinely required.<\/p>\n\n\n\n<h3 id=\"migration-tools\" class=\"wp-block-heading g36-review\">Preserve function-call state<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Tool workflows need intact call IDs, function names, thought signatures, and turn order. When using <code>generateContent<\/code>, each <code>FunctionResponse<\/code> should include its <code>call_id<\/code> \u548c <code>name<\/code>. Test malformed calls, explanatory text before tool output, and recovery from rejected or unavailable tools.<\/p>\n\n\n\n<ul class=\"wp-block-list g36-review g36-checklist\">\n<li>Change the target model ID to <code>gemini-3.6-flash<\/code>.<\/li>\n\n\n\n<li>\u79fb\u9664 <code>\u6eab\u5ea6<\/code>, <code>top_p<\/code>, <code>top_k<\/code>, \u4ee5\u53ca <code>candidate_count<\/code>.<\/li>\n\n\n\n<li>\u66ff\u63db <code>thinking_budget<\/code> \u8207 <code>\u601d\u8003\u5c64\u7d1a<\/code>.<\/li>\n\n\n\n<li>Remove model-turn prefilling.<\/li>\n\n\n\n<li>Preserve function call IDs, names, thought signatures, and tool order.<\/li>\n\n\n\n<li>Validate finish reasons, schemas, retries, and every success claim.<\/li>\n\n\n\n<li>Recalculate cost per successful task, including failed calls and tools.<\/li>\n\n\n\n<li>Run prompt-injection and destructive-action tests before Computer Use.<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"691\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/api-migration-1024x691.png\" alt=\"Migration summary: change request controls and preserve the complete tool trace. Confirm current API coverage before choosing Interactions.\" class=\"wp-image-16938\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/api-migration-1024x691.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/api-migration-300x203.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/api-migration-768x518.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/api-migration-1536x1037.png 1536w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/api-migration-18x12.png 18w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/api-migration.png 1600w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n<p><!-- \u4e0a\u4f20\u672c\u5730\u6587\u4ef6 output\/gemini-3-6-flash-wp\/visuals\/api-migration.png\uff1b\u5982\u5a92\u4f53\u5e93\u751f\u6210\u7684 URL \u4e0d\u540c\uff0c\u8bf7\u66ff\u6362\u4e0b\u65b9 src\u3002 --><\/p>\n\n\n<h2 id=\"comparison\" class=\"wp-block-heading g36-review\">Gemini 3.6 Flash vs 3.5 Flash and Flash-Lite<\/h2>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Gemini 3.6 Flash is the direct upgrade candidate for Gemini 3.5 Flash. Standard input pricing stays at $1.50 per million tokens, while output pricing falls from $9.00 to $7.50\u2014a 16.7% reduction before considering Google&#8217;s separate report of lower output-token use in one evaluation.<\/p>\n\n\n\n<figure class=\"g36-review wp-block-table g36-table g36-table--comparison\" tabindex=\"0\" role=\"region\" aria-label=\"Gemini 3.6 Flash versus 3.5 Flash comparison\" style=\"border:1px solid #d9e1ef;border-radius:16px;box-shadow:0 10px 28px rgba(18,38,82,.08);display:block;margin:24px 0 32px;max-width:100%;overflow-x:auto\"><table style=\"border-collapse:collapse;border-spacing:0;font-size:15px;margin:0;width:100%\"><thead><tr><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom\">\u985e\u5225<\/th><th class=\"g36-highlight\" style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom;background:#dceaff;color:#123d85;font-weight:850\">Gemini 3.6 \u9583\u5149\u71c8<\/th><th style=\"background:#172953;border:1px solid #33466f;color:#fff;font-weight:800;padding:13px 15px;text-align:left;vertical-align:bottom\">\u96d9\u5b50\u661f 3.5 Flash<\/th><\/tr><\/thead><tbody><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Standard input \/ 1M<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">$1.50<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">$1.50<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Standard output \/ 1M<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">$7.50<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">$9.00<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">DeepSWE v1.1<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">49.0%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">37.0%<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">MLE-Bench<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">63.9%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">49.7%<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">OSWorld-Verified<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">83.0%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">78.4%<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">1M GDM-MRCR v2<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">54.0%<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">26.6%<\/td><\/tr><tr><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">Our bounded code repair<\/td><td class=\"g36-highlight\" style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top;background:#dceaff;color:#123d85;font-weight:850\">5\/5 after repair<\/td><td style=\"background:#fff;border:1px solid #dce3ef;color:#293854;padding:12px 15px;text-align:left;vertical-align:top\">5\/5 after tolerant extraction<\/td><\/tr><\/tbody><\/table><figcaption style=\"background:#f5f7fc;border-top:1px solid #d9e1ef;color:#52617a;font-size:13px;line-height:1.5;margin:0;padding:12px 15px\">Official prices and benchmark values plus our separately labeled single-run coding fixture.<\/figcaption><\/figure>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The migration case is strongest when better agentic scores, lower output price, and shorter loops translate to more successful work per dollar. It is weakest when a specific repository or tool harness triggers incomplete execution, false verification, or repeated human correction. Run matched prompts with the same tools and acceptance tests.<\/p>\n\n\n\n<h3 id=\"flash-lite\" class=\"wp-block-heading g36-review\">Where Gemini 3.5 Flash-Lite fits<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Gemini 3.5 Flash-Lite is a lower-cost sibling for high-throughput extraction, classification, translation, routing, document processing, and bounded subagent work. Standard pricing is $0.30 per million input tokens and $2.50 per million output tokens. That makes 3.6 Flash five times more expensive on input and three times more expensive on output.<\/p>\n\n\n<p><!-- INTERNAL LINK TODO: add the site's Gemini 3.5 Flash-Lite review in this comparison after its URL is confirmed. --><\/p>\n\n\n<p class=\"g36-review wp-block-paragraph\">A two-tier router can default eligible work to Flash-Lite and escalate low-confidence, tool-heavy, multimodal, or long-horizon tasks to 3.6 Flash. The router should be observable and reversible, with reasons logged and both retry cost and correction cost included. Cheaper tokens do not automatically mean cheaper completed tasks.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"730\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/model-router-1024x730.png\" alt=\"A practical router starts lean and escalates on evidence such as low confidence, tool need, or long task horizon.\" class=\"wp-image-16939\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/model-router-1024x730.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/model-router-300x214.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/model-router-768x547.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/model-router-1536x1094.png 1536w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/model-router-18x12.png 18w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/model-router.png 1600w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n<p><!-- \u4e0a\u4f20\u672c\u5730\u6587\u4ef6 output\/gemini-3-6-flash-wp\/visuals\/model-router.png\uff1b\u5982\u5a92\u4f53\u5e93\u751f\u6210\u7684 URL \u4e0d\u540c\uff0c\u8bf7\u66ff\u6362\u4e0b\u65b9 src\u3002 --><\/p>\n\n\n<h2 id=\"early-feedback\" class=\"wp-block-heading g36-review\">Early user feedback and third-party results<\/h2>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Gemini 3.6 Flash had been public for roughly one day when these screenshots were captured on July 22, 2026. The posts are <strong>Social anecdote<\/strong> evidence, not a representative survey, and platform rankings are <strong>Third-party benchmark<\/strong> evidence rather than Google results or our measurements. They are useful for identifying hypotheses to test.<\/p>\n\n\n\n<h3 id=\"third-party-benchmarks\" class=\"wp-block-heading g36-review\">Promising browser and front-end signals, with caveats<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Arena.ai reported Gemini 3.6 Flash at number 12 with 1,537 points in Frontend Code Arena, up from number 21 for Gemini 3.5 Flash. The result reflects Arena&#8217;s prompts, voters, model snapshot, and ranking method. It supports a front-end improvement hypothesis but does not establish universal code quality.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-large\"><img alt=\"\" loading=\"lazy\" decoding=\"async\" width=\"321\" height=\"1024\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/x-02-arena-frontend-improvement-321x1024.png\" class=\"wp-image-16940\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/x-02-arena-frontend-improvement-321x1024.png 321w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/x-02-arena-frontend-improvement-94x300.png 94w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/x-02-arena-frontend-improvement-768x2447.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/x-02-arena-frontend-improvement-482x1536.png 482w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/x-02-arena-frontend-improvement-4x12.png 4w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/x-02-arena-frontend-improvement-scaled.png 803w\" sizes=\"(max-width: 321px) 100vw, 321px\" \/><\/figure>\n\n\n<p><!-- \u4e0a\u4f20\u672c\u5730\u6587\u4ef6 output\/gemini-3-6-flash-social-reviews\/x-02-arena-frontend-improvement.png\uff1b\u5982\u5a92\u4f53\u5e93\u751f\u6210\u7684 URL \u4e0d\u540c\uff0c\u8bf7\u66ff\u6362\u4e0b\u65b9 src\u3002 --><\/p>\n\n\n<p class=\"g36-review wp-block-paragraph\">Browser Use reported 68% on its BU Benchmark and called the model its best price-to-performance browser-agent option. This is a vendor-authored benchmark, not a neutral industry evaluation. It is directionally interesting alongside Google&#8217;s OSWorld result, but teams should reproduce browser tasks inside their own safety and tool-execution harness.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">A critical X user read Google&#8217;s table differently, arguing that the most convincing gains appeared in vision and context rather than coding. Another user pointed to a composite coding index where 3.6 sat slightly below 3.5. Different task mixes can produce apparently conflicting outcomes, which is why benchmark names and methods should remain attached to every number.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"822\" height=\"987\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/x-03-user-critical-benchmark-take.png\" alt=\"A critical reading of the official table from @synthwavedd. View the original X post. Commentary, not a hands-on evaluation.\" class=\"wp-image-16941\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/x-03-user-critical-benchmark-take.png 822w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/x-03-user-critical-benchmark-take-250x300.png 250w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/x-03-user-critical-benchmark-take-768x922.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/x-03-user-critical-benchmark-take-10x12.png 10w\" sizes=\"(max-width: 822px) 100vw, 822px\" \/><\/figure>\n\n\n<p><!-- \u4e0a\u4f20\u672c\u5730\u6587\u4ef6 output\/gemini-3-6-flash-social-reviews\/x-03-user-critical-benchmark-take.png\uff1b\u5982\u5a92\u4f53\u5e93\u751f\u6210\u7684 URL \u4e0d\u540c\uff0c\u8bf7\u66ff\u6362\u4e0b\u65b9 src\u3002 --><\/p>\n\n\n<h3 id=\"coding-anecdotes\" class=\"wp-block-heading g36-review\">Coding reports emphasize verification and completion<\/h3>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The most detailed Reddit report described 3.6 Flash as fast and capable when tasks were tightly scoped, measurable, and instrumented before repair. The same author warned about confident verification claims, destructive-looking autonomy, and the supervision cost of checking work. The author also disclosed that another model reviewed or fixed the result and that a setup issue confounded part of the test.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"665\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/reddit-01-antigravity-coding-review-post-1024x665.png\" alt=\"Mixed coding-agent report from u\/Valuable_Elevator948. \" class=\"wp-image-16942\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/reddit-01-antigravity-coding-review-post-1024x665.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/reddit-01-antigravity-coding-review-post-300x195.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/reddit-01-antigravity-coding-review-post-768x499.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/reddit-01-antigravity-coding-review-post-18x12.png 18w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/reddit-01-antigravity-coding-review-post.png 1354w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><a href=\"https:\/\/www.reddit.com\/r\/google_antigravity\/comments\/1v312ut\/gemini_36_flash_review\/\">View the original Reddit thread<\/a>. Early anecdote with disclosed confounders.<\/figcaption><\/figure>\n\n\n<p><!-- \u4e0a\u4f20\u672c\u5730\u6587\u4ef6 output\/gemini-3-6-flash-social-reviews\/reddit-01-antigravity-coding-review-post.png\uff1b\u5982\u5a92\u4f53\u5e93\u751f\u6210\u7684 URL \u4e0d\u540c\uff0c\u8bf7\u66ff\u6362\u4e0b\u65b9 src\u3002 --><\/p>\n\n\n<p class=\"g36-review wp-block-paragraph\">A separate Reddit author reported incomplete project work followed by a claim that the task was finished. Commenters described similar context, debugging, and hallucinated-verification concerns. The tool configuration and prompts were not independently verified, so the report should become a test case\u2014not a generalized verdict.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"431\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/reddit-03-bard-negative-coding-comments-1024x431.png\" alt=\"Negative project-coding report from u\/Embarrassed_Time_129. \" class=\"wp-image-16944\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/reddit-03-bard-negative-coding-comments-1024x431.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/reddit-03-bard-negative-coding-comments-300x126.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/reddit-03-bard-negative-coding-comments-768x323.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/reddit-03-bard-negative-coding-comments-18x8.png 18w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/07\/reddit-03-bard-negative-coding-comments.png 1350w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><a href=\"https:\/\/www.reddit.com\/r\/Bard\/comments\/1v320qs\/i_dont_know_where_the_gemini_36_flash_is_better\/\">View the original Reddit thread<\/a>. Anecdotal and workflow-specific.<\/figcaption><\/figure>\n\n\n<p><!-- \u4e0a\u4f20\u672c\u5730\u6587\u4ef6 output\/gemini-3-6-flash-social-reviews\/reddit-03-bard-negative-coding-post.png\uff1b\u5982\u5a92\u4f53\u5e93\u751f\u6210\u7684 URL \u4e0d\u540c\uff0c\u8bf7\u66ff\u6362\u4e0b\u65b9 src\u3002 --><\/p>\n\n\n<p class=\"g36-review wp-block-paragraph\">The common evaluation lesson is to separate patch quality from verification honesty. A model may suggest a useful change yet use a test that cannot distinguish the fix, omit the final task segment, or misstate remaining failures. Acceptance criteria should check the artifact, the test&#8217;s discriminatory power, the final state, and the accuracy of the completion report.<\/p>\n\n\n\n<h2 id=\"limitations\" class=\"wp-block-heading g36-review\">Limitations, safety, and evidence boundaries<\/h2>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Gemini 3.6 Flash retains foundation-model failure modes, including hallucinated facts, incomplete execution, and confident but unsupported conclusions. The model card also notes occasional slowness or timeout behavior. Its March 2026 knowledge cutoff means newer facts require grounding, a current trusted source, or supplied documents.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Google reports broadly positive safety comparisons with Gemini 3.5 Flash, including improved text-to-text and multilingual safety, unchanged image-to-text safety, and slight regressions in refusal tone and unjustified refusals. The model remained below Google&#8217;s Critical Capability Levels after additional testing. Those are publisher-reported evaluations, not independent guarantees for a particular deployment.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Computer Use requires stronger operational controls than chat. Screenshot-based agents can encounter prompt injection, deceptive controls, pop-ups, and irreversible actions. Use isolated environments, least-privilege credentials, domain and action allowlists, confirmations, audit logs, spending limits, and rollback plans.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Our browser test did not exercise Google&#8217;s native Computer Use endpoint and cannot support a native latency, safety, or success-rate claim. It tested screenshot interpretation and a local text-based action resolver. Likewise, the Broly PDF conversion error applies only to that gateway request path.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Our latency figures include Broly gateway routing and came from one small suite, not repeated distributions. We did not measure TTFT. Costs are estimates derived from reported usage and official Standard rates, not invoices, and the suite did not use Search Grounding.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Official benchmarks, third-party leaderboards, our gateway tests, and social anecdotes answer different questions. Keeping their labels visible prevents false precision: Google results describe Google&#8217;s evaluation, our records describe one reproducible route and fixture set, and public posts describe individual experiences.<\/p>\n\n\n\n<h2 id=\"who-should-use\" class=\"wp-block-heading g36-review\">Who should use Gemini 3.6 Flash?<\/h2>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Gemini 3.6 Flash is a strong candidate for teams already using Gemini 3.5 Flash, especially when output cost, multimodal input, tool calling, or agentic planning is material. It also fits new applications that can define explicit success criteria and supervise risk-sensitive actions.<\/p>\n\n\n\n<ul class=\"wp-block-list g36-review\">\n<li>Coding agents with bounded tasks, isolated workspaces, and discriminating automated tests.<\/li>\n\n\n\n<li>Chart, image, video, audio, and PDF analysis pipelines that produce text or structured data.<\/li>\n\n\n\n<li>Supervised browser or desktop agents with confirmation gates and local validation.<\/li>\n\n\n\n<li>Knowledge-work systems that need function calling, file search, caching, or grounded search.<\/li>\n\n\n\n<li>Teams able to compare cost per successful task against 3.5 Flash and other providers.<\/li>\n<\/ul>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">It is a weaker fit when the task can be handled reliably by Flash-Lite, when real-time voice requires the Live API, or when native image or audio generation is required. It is also a poor fit for unsupervised high-risk automation without rollback and independent verification.<\/p>\n\n\n\n<ul class=\"wp-block-list g36-review\">\n<li>Route simple classification and extraction to a lower-cost tier where quality holds.<\/li>\n\n\n\n<li>Do not assume a 1M context window replaces retrieval engineering.<\/li>\n\n\n\n<li>Do not accept self-reported \u201cverified\u201d status without a test that could fail.<\/li>\n\n\n\n<li>Do not generalize Broly gateway latency or PDF conversion behavior to native Google endpoints.<\/li>\n\n\n\n<li>Do not enable irreversible Computer Use actions without explicit confirmation.<\/li>\n<\/ul>\n\n\n\n<h2 id=\"verdict\" class=\"wp-block-heading g36-review\">\u6700\u7d42\u88c1\u6c7a<\/h2>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">This Gemini 3.6 Flash review finds a strong production candidate, not an automatic migration. The model combines lower output pricing, a broad input and tool surface, and meaningful official gains on DeepSWE, MLE-Bench, OSWorld, and long-context retrieval. Our bounded tests add encouraging evidence for exact extraction, chart reading, focused code repair, and screenshot-guided local action planning.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">The caution is equally concrete. Tight output caps caused two initial truncations; reasoning used nearly the whole streaming completion budget; our 128K synthetic retrieval test failed twice; and the browser plan invented an unseen ID. The PDF path remained untested natively because the Broly conversion route failed before a model response.<\/p>\n\n\n\n<p class=\"g36-review wp-block-paragraph\">Adopt Gemini 3.6 Flash when a matched evaluation shows better completed-task economics than your current model. Track pass rate, tokens, retries, tool calls, unwanted edits, human correction time, and whether every verification claim is backed by a discriminating test. For mixed workloads, start with Flash-Lite and escalate on observable need.<\/p>\n\n\n<p><!-- INTERNAL LINK TODO: add the site's AI-agent evaluation checklist as the closing next step when its URL is available. --><\/p>\n\n\n<aside class=\"g36-review g36-cta\" aria-label=\"Recommended next step\" style=\"background:linear-gradient(135deg,#eef4ff,#f4f0ff 72%,#fff7fd);border:1px solid #cfc5ec;border-left:6px solid #6949b8;border-radius:16px;box-shadow:0 10px 28px rgba(18,38,82,.08);color:#323858;margin:24px 0 32px;padding:20px 22px\"><p style=\"margin:0\"><strong>\u4e0b\u4e00\u6b65\uff1a<\/strong> run the same bounded task suite against Gemini 3.6 Flash and your current production model. Choose on total cost per verified success\u2014not list price, one benchmark, or a model&#8217;s own claim that the work is done.<\/p><\/aside>\n\n\n\n<h2 id=\"faq\" class=\"wp-block-heading g36-review\">\u5e38\u898b\u554f\u984c<\/h2>\n\n\n\n<h3 id=\"faq-ga\" class=\"wp-block-heading g36-review\">Is Gemini 3.6 Flash generally available?<\/h3>\n\n\n\n<p class=\"g36-review g36-faq-answer wp-block-paragraph\">Yes. Google lists Gemini 3.6 Flash as stable and generally available for production use. The model ID is <code>gemini-3.6-flash<\/code>, and it is accessible through Google AI Studio and the Gemini Developer API. API feature coverage still varies between Interactions and <code>generateContent<\/code>.<\/p>\n\n\n\n<h3 id=\"faq-cost\" class=\"wp-block-heading g36-review\">How much does Gemini 3.6 Flash cost?<\/h3>\n\n\n\n<p class=\"g36-review g36-faq-answer wp-block-paragraph\">Standard paid pricing is $1.50 per million input tokens and $7.50 per million output tokens, including thinking tokens. Batch and Flex cost $0.75 input and $3.75 output, while Priority costs $2.70 input and $13.50 output. Caching and grounding can add separate charges.<\/p>\n\n\n\n<h3 id=\"faq-context\" class=\"wp-block-heading g36-review\">What is the Gemini 3.6 Flash context window?<\/h3>\n\n\n\n<p class=\"g36-review g36-faq-answer wp-block-paragraph\">Gemini 3.6 Flash supports up to 1,048,576 input tokens and 65,536 output tokens. That is a capacity limit, not a retrieval guarantee. Google&#8217;s 1M GDM-MRCR result was 54.0%, and our different synthetic 128K gateway task failed, so critical systems still need retrieval and validation.<\/p>\n\n\n\n<h3 id=\"faq-modalities\" class=\"wp-block-heading g36-review\">Does Gemini 3.6 Flash support images, audio, video, and PDFs?<\/h3>\n\n\n\n<p class=\"g36-review g36-faq-answer wp-block-paragraph\">Yes, Google lists text, images, audio, video, and PDFs as supported inputs. The model produces text, structured data, and tool calls rather than native images or audio. Our Broly PDF conversion path was unsupported, but that gateway result does not establish a native Gemini PDF limitation.<\/p>\n\n\n\n<h3 id=\"faq-computer-use\" class=\"wp-block-heading g36-review\">Does Gemini 3.6 Flash support Computer Use?<\/h3>\n\n\n\n<p class=\"g36-review g36-faq-answer wp-block-paragraph\">Yes. Google&#8217;s Gemini API supports Computer Use with Gemini 3.6 Flash for browser, mobile, and desktop environments, but the feature remains Preview. The client executes proposed actions and must apply confirmations and safety controls. Our local screenshot harness was not a native Computer Use test.<\/p>\n\n\n\n<h3 id=\"faq-vs-35\" class=\"wp-block-heading g36-review\">Is Gemini 3.6 Flash better than Gemini 3.5 Flash?<\/h3>\n\n\n\n<p class=\"g36-review g36-faq-answer wp-block-paragraph\">It is stronger on several Google comparisons and cheaper per output token, but \u201cbetter\u201d depends on the workload. Our bounded coding fixture ended 5\/5 for both models, while 3.5 needed tolerant code extraction. Run matched tasks and compare verified completion cost, not model names alone.<\/p>\n\n\n\n<h3 id=\"faq-sampling\" class=\"wp-block-heading g36-review\">Are temperature, top_p, and top_k supported?<\/h3>\n\n\n\n<p class=\"g36-review g36-faq-answer wp-block-paragraph\">Google says <code>\u6eab\u5ea6<\/code>, <code>top_p<\/code>, \u4ee5\u53ca <code>top_k<\/code> are deprecated and ignored for these models, with future model generations expected to reject them. Remove the fields during migration. Use clear instructions, schemas, structured outputs, and application validators to control behavior.<\/p>\n\n\n\n<h3 id=\"faq-free\" class=\"wp-block-heading g36-review\">Is there a free tier?<\/h3>\n\n\n\n<p class=\"g36-review g36-faq-answer wp-block-paragraph\">Google lists free Standard-tier input and output access subject to rate limits and availability. The pricing page says free-tier content may be used to improve Google products, while paid-tier content is marked otherwise. Review current rate, data-use, and regional terms before sending production or sensitive data.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>","protected":false},"excerpt":{"rendered":"<p>At a glance Gemini 3.6 Flash is Google&#8217;s stable, natively multimodal workhorse model for text generation, coding, document analysis, and tool-using agents. It accepts text, images, audio, video, and PDFs, supports a 1,048,576-token input window, and produces text or tool calls rather than native image or audio output. Gemini 3.6 Flash is worth testing for [&hellip;]<\/p>","protected":false},"author":16,"featured_media":16948,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_seopress_robots_primary_cat":"","_seopress_titles_title":"Gemini 3.6 Flash Review: Pricing, Benchmarks and API","_seopress_titles_desc":"Gemini 3.6 Flash review with official pricing, benchmarks, API migration guidance, and verified hands-on Broly gateway tests.","_seopress_robots_index":"","footnotes":""},"categories":[7],"tags":[],"class_list":["post-16724","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-chat"],"_links":{"self":[{"href":"https:\/\/wp.glbgpt.com\/zh-hk\/wp-json\/wp\/v2\/posts\/16724","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wp.glbgpt.com\/zh-hk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wp.glbgpt.com\/zh-hk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wp.glbgpt.com\/zh-hk\/wp-json\/wp\/v2\/users\/16"}],"replies":[{"embeddable":true,"href":"https:\/\/wp.glbgpt.com\/zh-hk\/wp-json\/wp\/v2\/comments?post=16724"}],"version-history":[{"count":7,"href":"https:\/\/wp.glbgpt.com\/zh-hk\/wp-json\/wp\/v2\/posts\/16724\/revisions"}],"predecessor-version":[{"id":16949,"href":"https:\/\/wp.glbgpt.com\/zh-hk\/wp-json\/wp\/v2\/posts\/16724\/revisions\/16949"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wp.glbgpt.com\/zh-hk\/wp-json\/wp\/v2\/media\/16948"}],"wp:attachment":[{"href":"https:\/\/wp.glbgpt.com\/zh-hk\/wp-json\/wp\/v2\/media?parent=16724"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wp.glbgpt.com\/zh-hk\/wp-json\/wp\/v2\/categories?post=16724"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wp.glbgpt.com\/zh-hk\/wp-json\/wp\/v2\/tags?post=16724"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}