{"id":20658,"date":"2026-10-14T07:32:43","date_gmt":"2026-10-14T11:32:43","guid":{"rendered":"https:\/\/wp.glbgpt.com\/?p=20658"},"modified":"2026-10-14T07:32:44","modified_gmt":"2026-10-14T11:32:44","slug":"muse-wonder-review","status":"publish","type":"post","link":"https:\/\/wp.glbgpt.com\/fr\/hub\/muse-wonder-review","title":{"rendered":"Avis sur Muse Wonder : tests r\u00e9els, tarifs et limites"},"content":{"rendered":"<p class=\"wp-block-paragraph\"><strong>Muse Wonder is worth trying for code repair and source-based analysis.<\/strong> The examples below passed 29 independent checks: 13 for the code-repair follow-up and eight each for source reconciliation and recovery planning. With explicit identity requirements, the returned code preserved distinct Symbols and object references while keeping the requested order and format.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">M\u00e9ta <a href=\"https:\/\/research.meta.ai\/blog\/muse-wonder\">introduced Muse Wonder on October 12<\/a> as a frontier model for personal and professional agents. The bigger promises involve long tasks, memory and computer use. Our text tests give an early view of how it handles constraints; they do not settle those bigger claims.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">You can repeat the tasks through <a href=\"https:\/\/www.glbgpt.com\/home\/muse-wonder?inviter=hub_content_musewonder&amp;login=1\">Muse Wonder on GlobalGPT<\/a> before changing your workflow. This review is published by GlobalGPT, and the platform entry is also where we ran the tests.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-large\"><img alt=\"\" fetchpriority=\"high\" decoding=\"async\" width=\"1024\" height=\"547\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/image-2-1024x547.png\" class=\"wp-image-20659\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/image-2-1024x547.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/image-2-300x160.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/image-2-768x410.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/image-2-18x10.png 18w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/image-2-1536x820.png 1536w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/image-2-2048x1094.png 2048w\" 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-text-color has-background has-link-color wp-element-button\" href=\"https:\/\/www.glbgpt.com\/home\/muse-wonder?inviter=hub_content_musewonder&amp;login=1\" style=\"background:linear-gradient(63deg,rgb(6,147,227) 75%,rgb(155,81,224) 100%)\"><strong>Try on GlobalGPT<\/strong><\/a><\/div>\n<\/div>\n\n\n\n<nav aria-label=\"Table des mati\u00e8res\" style=\"border:1px solid #d4e1df;border-radius:17px;padding:20px;margin:22px 0;background:linear-gradient(130deg,#163d43,#30496e);color:#fff\"><p style=\"margin:0 0 9px;font:700 12px\/1.5 system-ui;letter-spacing:.1em;text-transform:uppercase;color:#93cce2\">Inside this review<\/p><div style=\"display:grid;grid-template-columns:repeat(auto-fit,minmax(min(100%,210px),1fr));gap:12px;\"><a href=\"#what-is-muse-wonder\" style=\"color:#fff;text-decoration:none;padding:6px 0;font-size:14px;line-height:1.5\">01 \u00b7 What is Muse Wonder?<\/a><a href=\"#capabilities\" style=\"color:#fff;text-decoration:none;padding:6px 0;font-size:14px;line-height:1.5\">02 \u00b7 What Meta promises: memory, focus and context<\/a><a href=\"#hands-on\" style=\"color:#fff;text-decoration:none;padding:6px 0;font-size:14px;line-height:1.5\">03 \u00b7 Three real Muse Wonder tests and results<\/a><a href=\"#social-reaction\" style=\"color:#fff;text-decoration:none;padding:6px 0;font-size:14px;line-height:1.5\">04 \u00b7 X benchmarks and Reddit reactions<\/a><a href=\"#pricing\" style=\"color:#fff;text-decoration:none;padding:6px 0;font-size:14px;line-height:1.5\">05 \u00b7 Muse Wonder pricing: tokens versus subscriptions<\/a><a href=\"#access\" style=\"color:#fff;text-decoration:none;padding:6px 0;font-size:14px;line-height:1.5\">06 \u00b7 How to use Muse Wonder on GlobalGPT<\/a><a href=\"#verdict\" style=\"color:#fff;text-decoration:none;padding:6px 0;font-size:14px;line-height:1.5\">07 \u00b7 Who should use it? Our verdict<\/a><a href=\"#faq\" style=\"color:#fff;text-decoration:none;padding:6px 0;font-size:14px;line-height:1.5\">08 \u00b7 Muse Wonder review FAQ<\/a><\/div><\/nav>\n\n\n\n<h2 id=\"what-is-muse-wonder\" class=\"wp-block-heading\">What is Muse Wonder?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Le <a href=\"https:\/\/research.meta.ai\/blog\/muse-wonder\">official Meta launch announcement<\/a> introduces two variants: Muse Wonder and Muse Wonder Full. It says Wonder is available through Meta Model API and Muse Code, while Full is available to Muse Code subscribers. It separately says Wonder will soon power the consumer Muse product, starting with subscribers. Those are three access surfaces with different scopes.<\/p>\n\n\n\n<figure style=\"margin:23px 0\"><a href=\"https:\/\/research.meta.ai\/blog\/muse-wonder\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/static.futureshareai.com\/glb_features\/mcp\/4\/muse-wonder-official-launch-excerpt_136b5c77ce26494887dc1f51f0748aac.webp\" alt=\"Meta AI Research launch excerpt describing Muse Wonder, Active Memory, Active Focus, API access and the Muse Code-only Full variant.\" width=\"1280\" height=\"720\" style=\"display:block;max-width:100%;height:auto;border:1px solid #d5dcd7;border-radius:12px\"><\/a><figcaption style=\"font-size:13px;line-height:1.6;color:#546a66;margin-top:9px\">Meta\u2019s launch statement separates Wonder API access, Wonder Full in Muse Code and the staged Muse consumer rollout. Captured October 14, 2026. Source: <a href=\"https:\/\/research.meta.ai\/blog\/muse-wonder\">Meta AI Research<\/a>.<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Meta\u2019s <a href=\"https:\/\/dev.meta.ai\/docs\/models\">documentation du mod\u00e8le<\/a> identifies the standard preview model as <code>muse-wonder-1.0<\/code>. It accepts text, images, video, audio and PDFs, and returns text. Multimodal understanding can help a model explain a file or reason about a screen; it does not by itself give that model an image- or video-generation endpoint.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Choose an evaluation around what you actually need: correct code, a cited answer, a usable plan or a finished tool task. Our <a href=\"https:\/\/www.glbgpt.com\/hub\/best-ai-models\/\">Guide des meilleurs mod\u00e8les d'IA<\/a> helps separate those workloads before you turn a model name into a buying decision.<\/p>\n\n\n\n<h2 id=\"capabilities\" class=\"wp-block-heading\">What Meta promises: memory, focus and context<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Le <a href=\"https:\/\/dev.meta.ai\/docs\/models\">documented context window is 1,048,576 tokens<\/a>. Meta also describes tool calling and structured outputs. A large window is useful when a task has many files, but the specification alone says little about retrieval errors, conflicting instructions or what survives a long interaction.<\/p>\n\n\n\n<div style=\"display:grid;grid-template-columns:repeat(auto-fit,minmax(min(100%,210px),1fr));gap:12px;margin:20px 0\"><div style=\"border-top:4px solid #577aa4;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">Contexte<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">1,048,576<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">Tokens documented by Meta; not stress-tested here.<\/p><\/div><div style=\"border-top:4px solid #8c75b6;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">Entr\u00e9e<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">Multimodal<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">Text, images, video, audio and PDFs.<\/p><\/div><div style=\"border-top:4px solid #cc8c57;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">Sortie<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">Texte<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">Plans, analysis and code; our runs used text only.<\/p><\/div><div style=\"border-top:4px solid #4b937a;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">Acc\u00e8s<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">Two variants<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">Standard Wonder and Muse Code\u2019s Wonder Full.<\/p><\/div><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">In Meta\u2019s <a href=\"https:\/\/x.com\/AIatMeta\/status\/2109691156258783589\">Active Memory and Active Focus launch post<\/a>, Memory is about carrying relevant context across conversations; Focus is about staying with a goal when a task becomes difficult. These are attractive ideas for a working agent. Testing them needs repeated conversations and actual tool execution, not a short prompt that merely asks the model to explain the features.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">That distinction shapes our test set. We tested a coding constraint, an evidence conflict and a recovery plan. The source packet fits comfortably into an ordinary chat. We did not attach images or PDFs, fill the context window, or measure persistent memory between chats.<\/p>\n\n\n\n<h2 id=\"hands-on\" class=\"wp-block-heading\">Three real Muse Wonder tests and results<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">We tested three task types in GlobalGPT\u2019s production Muse Wonder interface on October 14, 2026, using its default settings. The code example below is a follow-up with explicit identity requirements; the source and planning examples retain their original outputs. The published entry maps to agent 6256 and label <code>muse-wonder<\/code>. No reasoning-effort control was exposed, and the provider backend version was not independently attested; these are not Wonder Full or max-effort API benchmark results.<\/p>\n\n\n\n<div style=\"display:grid;grid-template-columns:repeat(auto-fit,minmax(min(100%,210px),1fr));gap:12px;margin:20px 0\"><div style=\"border-top:4px solid #4b937a;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">Code basics<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">8 \/ 8<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">Order, types, missing IDs and input preservation passed.<\/p><\/div><div style=\"border-top:4px solid #577aa4;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">Evidence JSON<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">8 \/ 8<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">Format, source choice and missing facts passed.<\/p><\/div><div style=\"border-top:4px solid #8c75b6;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">Recovery plan<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">8 \/ 8<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">Six steps, 126 words and a bounded retry.<\/p><\/div><div style=\"border-top:4px solid #4b937a;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">ID identity<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">5 \/ 5<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">Symbols, references and BigInt passed in the follow-up.<\/p><\/div><\/div>\n\n\n\n<h3 class=\"wp-block-heading\">Test 1: fix a real JavaScript deduplication bug<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The original function used <code>record.id || ''<\/code>, which collapses numeric zero, false and absent IDs into the same value. We tested stable order, case-insensitive strings, separate missing-ID records and no mutation, then followed up with explicit Symbol, object-reference and BigInt requirements. The complete prompt and returned code below are from that follow-up.<\/p>\n\n\n\n<section id=\"t1-prompt\" data-text-card=\"copy\" style=\"border:1px solid #d4e1df;border-radius:17px;padding:20px;margin:22px 0;background:#132d32;color:#edf6f5;min-width:0\"><p style=\"margin:0 0 9px;font:700 12px\/1.5 system-ui;letter-spacing:.1em;text-transform:uppercase;color:#93cce2\">Entr\u00e9e exacte<\/p><div style=\"display:flex;flex-wrap:wrap;gap:10px;align-items:center;justify-content:space-between;margin-bottom:14px\"><strong style=\"font:700 19px\/1.4 Georgia,serif\">JavaScript repair \u00b7 complete prompt<\/strong><button type=\"button\" data-copy onclick=\"var c=this.closest('[data-text-card]'),e=c.querySelector('[data-value]'),s=c.querySelector('[role=status]'),v=e.textContent;var fallback=function(){e.focus();var r=document.createRange();r.selectNodeContents(e);var sel=window.getSelection();sel.removeAllRanges();sel.addRange(r);s.textContent='Selected all. Press Ctrl\/Cmd+C.';};try{if(!navigator.clipboard||!navigator.clipboard.writeText){fallback();return;}Promise.race([navigator.clipboard.writeText(v),new Promise(function(_,reject){setTimeout(function(){reject(new Error('timeout'));},900);})]).then(function(){s.textContent='Copied.';},fallback);}catch(_){fallback();}\" style=\"border:1px solid #93b7aa;background:#d3e6cc;color:#183a32;border-radius:8px;padding:9px 15px;font:700 13px\/1.4 system-ui;cursor:pointer\">Copier l'invite<\/button><\/div><pre data-value tabindex=\"0\" contenteditable=\"true\" spellcheck=\"false\" role=\"textbox\" aria-multiline=\"true\" aria-label=\"JavaScript repair \u00b7 complete prompt\" style=\"margin:0;padding:15px;border-radius:10px;border:1px solid #45636a;background:#0c2026;color:#edf6f5;font:13px\/1.75 ui-monospace,monospace;white-space:pre-wrap;overflow-wrap:anywhere;min-width:0\">Follow-up code-repair test for the same uniqueById task. Return a corrected runnable JavaScript function, a short explanation, and exactly five assert-style tests. Do not claim you executed tests.\n\nAcceptance requirements:\n- Preserve input order and return the first record for each duplicate ID. Never mutate the array or records.\n- Records whose id is absent, null or undefined are always kept.\n- Compare string IDs case-insensitively. Keep number 0, boolean false and string '0' distinct.\n- Preserve JavaScript ID identity: two distinct Symbol('x') values must remain separate, repeating the very same Symbol must deduplicate, and repeated Symbol.for('shared') values must deduplicate.\n- Different objects or functions used as IDs remain distinct; repeated references deduplicate. BigInt 1n and number 1 remain distinct.\n- Use the original sample [{id:0},{id:false},{id:'A'},{id:'a'},{name:'one'},{name:'two'}], keeping 0, false, first A, and both id-less records.\n\nThe five supplied tests should cover the ordinary sample, missing\/null IDs, Symbol identity, object-reference identity and BigInt versus number. Use an implementation that preserves values rather than collapsing distinct identities into textual keys.<\/pre><span role=\"status\" aria-live=\"polite\" style=\"display:block;margin-top:10px;font:12px\/1.5 system-ui;color:#b9d3d1\">If automatic copy is blocked, select the text and press Ctrl\/Cmd+C.<\/span><\/section>\n\n\n\n<section id=\"t1-function\" data-text-card=\"copy\" style=\"border:1px solid #d4e1df;border-radius:17px;padding:20px;margin:22px 0;background:#132d32;color:#edf6f5;min-width:0\"><p style=\"margin:0 0 9px;font:700 12px\/1.5 system-ui;letter-spacing:.1em;text-transform:uppercase;color:#93cce2\">R\u00e9sultats du mod\u00e8le<\/p><div style=\"display:flex;flex-wrap:wrap;gap:10px;align-items:center;justify-content:space-between;margin-bottom:14px\"><strong style=\"font:700 19px\/1.4 Georgia,serif\">Returned function \u00b7 unchanged<\/strong><button type=\"button\" data-copy onclick=\"var c=this.closest('[data-text-card]'),e=c.querySelector('[data-value]'),s=c.querySelector('[role=status]'),v=e.textContent;var fallback=function(){e.focus();var r=document.createRange();r.selectNodeContents(e);var sel=window.getSelection();sel.removeAllRanges();sel.addRange(r);s.textContent='Selected all. Press Ctrl\/Cmd+C.';};try{if(!navigator.clipboard||!navigator.clipboard.writeText){fallback();return;}Promise.race([navigator.clipboard.writeText(v),new Promise(function(_,reject){setTimeout(function(){reject(new Error('timeout'));},900);})]).then(function(){s.textContent='Copied.';},fallback);}catch(_){fallback();}\" style=\"border:1px solid #93b7aa;background:#d3e6cc;color:#183a32;border-radius:8px;padding:9px 15px;font:700 13px\/1.4 system-ui;cursor:pointer\">Copier le code<\/button><\/div><pre data-value tabindex=\"0\" contenteditable=\"true\" spellcheck=\"false\" role=\"textbox\" aria-multiline=\"true\" aria-label=\"Returned function \u00b7 unchanged\" style=\"margin:0;padding:15px;border-radius:10px;border:1px solid #45636a;background:#0c2026;color:#edf6f5;font:13px\/1.75 ui-monospace,monospace;white-space:pre-wrap;overflow-wrap:anywhere;min-width:0\">function uniqueById(records) {\n  const seen = new Set();\n\n  return records.filter((record) =&gt; {\n    \/\/ Absent, null, and undefined IDs are always kept.\n    if (record == null || record.id == null) {\n      return true;\n    }\n\n    const id = record.id;\n\n    \/\/ Normalize only strings for case-insensitive comparison.\n    \/\/ All other values are stored as-is, preserving their identity.\n    const key = typeof id === &quot;string&quot; ? id.toLowerCase() : id;\n\n    if (seen.has(key)) {\n      return false;\n    }\n\n    seen.add(key);\n    return true;\n  });\n}<\/pre><span role=\"status\" aria-live=\"polite\" style=\"display:block;margin-top:10px;font:12px\/1.5 system-ui;color:#b9d3d1\">If automatic copy is blocked, select the text and press Ctrl\/Cmd+C.<\/span><\/section>\n\n\n\n<p class=\"wp-block-paragraph\">Muse Wonder lowercased only string IDs and stored other IDs directly in a <code>D\u00e9finir<\/code>. Records with absent, <code>null<\/code> ou <code>undefined<\/code> IDs always pass through. The eight ordinary checks covered the sample, empty input, case handling, distinct types, missing IDs, immutability, order and returned-record identity. All eight passed.<\/p>\n\n\n\n<section id=\"t1-model-tests\" data-text-card=\"result\" style=\"border:1px solid #d4e1df;border-radius:17px;padding:20px;margin:22px 0;background:#132d32;color:#edf6f5;min-width:0\"><p style=\"margin:0 0 9px;font:700 12px\/1.5 system-ui;letter-spacing:.1em;text-transform:uppercase;color:#93cce2\">R\u00e9sultats du mod\u00e8le<\/p><div style=\"display:flex;flex-wrap:wrap;gap:10px;align-items:center;justify-content:space-between;margin-bottom:14px\"><strong style=\"font:700 19px\/1.4 Georgia,serif\">Five model-supplied tests \u00b7 executed and passed<\/strong><\/div><pre data-value tabindex=\"0\" style=\"margin:0;padding:15px;border-radius:10px;border:1px solid #45636a;background:#0c2026;color:#edf6f5;font:13px\/1.75 ui-monospace,monospace;white-space:pre-wrap;overflow-wrap:anywhere;min-width:0\">const assert = require(&quot;node:assert\/strict&quot;);\n\n\/\/ Test 1: Ordinary sample \u2014 string case, distinct primitive types, id-less records.\nassert.deepStrictEqual(\n  uniqueById([\n    { id: 0 },\n    { id: false },\n    { id: &quot;A&quot; },\n    { id: &quot;a&quot; },\n    { name: &quot;one&quot; },\n    { name: &quot;two&quot; },\n  ]),\n  [\n    { id: 0 },\n    { id: false },\n    { id: &quot;A&quot; },\n    { name: &quot;one&quot; },\n    { name: &quot;two&quot; },\n  ]\n);\n\n\/\/ Test 2: Missing, null, and undefined IDs are always kept.\nassert.deepStrictEqual(\n  uniqueById([\n    { name: &quot;one&quot; },\n    { id: null },\n    { id: undefined },\n    { name: &quot;two&quot; },\n    { id: null },\n  ]),\n  [\n    { name: &quot;one&quot; },\n    { id: null },\n    { id: undefined },\n    { name: &quot;two&quot; },\n    { id: null },\n  ]\n);\n\n\/\/ Test 3: Symbol identity and registered symbols.\nconst symbolA = Symbol(&quot;x&quot;);\nconst symbolB = Symbol(&quot;x&quot;);\nassert.deepStrictEqual(\n  uniqueById([\n    { id: symbolA },\n    { id: symbolB },\n    { id: symbolA },\n    { id: Symbol.for(&quot;shared&quot;) },\n    { id: Symbol.for(&quot;shared&quot;) },\n  ]),\n  [{ id: symbolA }, { id: symbolB }, { id: Symbol.for(&quot;shared&quot;) }]\n);\n\n\/\/ Test 4: Object and function reference identity.\nconst objectA = {};\nconst objectB = {};\nfunction functionA() {}\nfunction functionB() {}\nassert.deepStrictEqual(\n  uniqueById([\n    { id: objectA },\n    { id: objectB },\n    { id: objectA },\n    { id: functionA },\n    { id: functionB },\n    { id: functionA },\n  ]),\n  [\n    { id: objectA },\n    { id: objectB },\n    { id: functionA },\n    { id: functionB },\n  ]\n);\n\n\/\/ Test 5: BigInt, number, and string IDs remain distinct.\nassert.deepStrictEqual(\n  uniqueById([\n    { id: 1n },\n    { id: 1 },\n    { id: &quot;1&quot; },\n    { id: 1n },\n    { id: 1 },\n    { id: &quot;1&quot; },\n  ]),\n  [{ id: 1n }, { id: 1 }, { id: &quot;1&quot; }]\n);<\/pre><\/section>\n\n\n\n<p class=\"wp-block-paragraph\">We executed the model\u2019s five assert-style tests and all passed. We also ran our own checks separately. Execution happened in our verification environment after receiving the answer; the model did not operate a repository or independently attest a test run.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Five additional identity checks passed: distinct same-description Symbols remain separate; repeated identical Symbols deduplicate; registered Symbols deduplicate; objects and functions preserve reference identity; and <code>1n<\/code> remains distinct from <code>1<\/code>. That makes 13 independent checks for this follow-up version, in addition to the five model-supplied assertions.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">If code quality is your main reason to switch, use the <a href=\"https:\/\/www.glbgpt.com\/hub\/best-ai-model-for-coding\/\">Meilleur mod\u00e8le d'IA pour la comparaison de codes<\/a> to choose a second model and run the same input. We did not run a matched competitor test here, so this result does not establish a ranking.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Test 2: reconcile conflicting evidence into valid JSON<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The second task used a fictional Orion Agent packet. An older announcement said $19\/month and EU availability; newer official sources said $24\/month and a paused EU rollout. An unsupported community claim said $12. The missing output price and restart date were deliberate traps for invented certainty. None of these fictional prices are Muse Wonder pricing.<\/p>\n\n\n\n<section id=\"t2-prompt\" data-text-card=\"copy\" style=\"border:1px solid #d4e1df;border-radius:17px;padding:20px;margin:22px 0;background:#132d32;color:#edf6f5;min-width:0\"><p style=\"margin:0 0 9px;font:700 12px\/1.5 system-ui;letter-spacing:.1em;text-transform:uppercase;color:#93cce2\">Entr\u00e9e exacte<\/p><div style=\"display:flex;flex-wrap:wrap;gap:10px;align-items:center;justify-content:space-between;margin-bottom:14px\"><strong style=\"font:700 19px\/1.4 Georgia,serif\">Conflicting sources \u00b7 complete prompt<\/strong><button type=\"button\" data-copy onclick=\"var c=this.closest('[data-text-card]'),e=c.querySelector('[data-value]'),s=c.querySelector('[role=status]'),v=e.textContent;var fallback=function(){e.focus();var r=document.createRange();r.selectNodeContents(e);var sel=window.getSelection();sel.removeAllRanges();sel.addRange(r);s.textContent='Selected all. Press Ctrl\/Cmd+C.';};try{if(!navigator.clipboard||!navigator.clipboard.writeText){fallback();return;}Promise.race([navigator.clipboard.writeText(v),new Promise(function(_,reject){setTimeout(function(){reject(new Error('timeout'));},900);})]).then(function(){s.textContent='Copied.';},fallback);}catch(_){fallback();}\" style=\"border:1px solid #93b7aa;background:#d3e6cc;color:#183a32;border-radius:8px;padding:9px 15px;font:700 13px\/1.4 system-ui;cursor:pointer\">Copier l'invite<\/button><\/div><pre data-value tabindex=\"0\" contenteditable=\"true\" spellcheck=\"false\" role=\"textbox\" aria-multiline=\"true\" aria-label=\"Conflicting sources \u00b7 complete prompt\" style=\"margin:0;padding:15px;border-radius:10px;border:1px solid #45636a;background:#0c2026;color:#edf6f5;font:13px\/1.75 ui-monospace,monospace;white-space:pre-wrap;overflow-wrap:anywhere;min-width:0\">Muse Wonder review test 2: evidence reconciliation.\nUse only this fictional source packet. Return ONLY valid JSON with keys launch_date, plan_monthly_usd, api_input_per_million_usd, region_status, unresolved, citations. Each factual value must have its source IDs in citations. Do not silently average contradictions or invent missing facts.\n\n[S1] Product announcement, dated 2026-10-12: Orion Agent launched October 12. Pro plan $19\/month. Available in US and EU.\n[S2] Official pricing page, updated 2026-10-14: Pro plan $24\/month, billed monthly. API input $4 per million tokens. API output is not listed.\n[S3] Official status page, updated 2026-10-14: US access operational. EU rollout paused; no restart date supplied.\n[S4] Community post, dated 2026-10-14: Someone says the plan is $12 and EU works; no screenshot or link.\n\nPrefer newer official evidence for the same field. List any unresolved information explicitly.<\/pre><span role=\"status\" aria-live=\"polite\" style=\"display:block;margin-top:10px;font:12px\/1.5 system-ui;color:#b9d3d1\">If automatic copy is blocked, select the text and press Ctrl\/Cmd+C.<\/span><\/section>\n\n\n\n<section id=\"t2-output\" data-text-card=\"result\" style=\"border:1px solid #d4e1df;border-radius:17px;padding:20px;margin:22px 0;background:#132d32;color:#edf6f5;min-width:0\"><p style=\"margin:0 0 9px;font:700 12px\/1.5 system-ui;letter-spacing:.1em;text-transform:uppercase;color:#93cce2\">R\u00e9sultats du mod\u00e8le<\/p><div style=\"display:flex;flex-wrap:wrap;gap:10px;align-items:center;justify-content:space-between;margin-bottom:14px\"><strong style=\"font:700 19px\/1.4 Georgia,serif\">Returned JSON \u00b7 original copied text<\/strong><\/div><pre data-value tabindex=\"0\" style=\"margin:0;padding:15px;border-radius:10px;border:1px solid #45636a;background:#0c2026;color:#edf6f5;font:13px\/1.75 ui-monospace,monospace;white-space:pre-wrap;overflow-wrap:anywhere;min-width:0\">{&quot;launch_date&quot;: &quot;2026-10-12&quot;, &quot;plan_monthly_usd&quot;: 24, &quot;api_input_per_million_usd&quot;: 4, &quot;region_status&quot;: {&quot;US&quot;: &quot;operational&quot;, &quot;EU&quot;: &quot;paused&quot;}, &quot;unresolved&quot;: [&quot;API output price is not listed in the sources.&quot;, &quot;EU rollout restart date is not supplied.&quot;, &quot;Conflicting claims were not used as the basis for the selected values: S1 listed the Pro plan at $19\/month and EU as available, while the unofficial and unsupported S4 claimed $12\/month and that EU works; the newer official S2 and S3 were preferred instead.&quot;], &quot;citations&quot;: {&quot;launch_date&quot;: [&quot;S1&quot;], &quot;plan_monthly_usd&quot;: [&quot;S2&quot;], &quot;api_input_per_million_usd&quot;: [&quot;S2&quot;], &quot;region_status&quot;: [&quot;S3&quot;]}}<\/pre><\/section>\n\n\n\n<p class=\"wp-block-paragraph\">The output parsed as JSON and contained the six requested keys. It chose 24 from S2, chose \u201cpaused\u201d from S3, retained S1 for the launch date and kept the missing output price and EU restart date unresolved. It did not cite S4 as the basis for a selected fact. All eight checks passed.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This is the most useful result for everyday research work: the model preferred newer official evidence for the same field without erasing what it still could not know. It is a small source-reconciliation test, not a web-search accuracy score or a million-token retrieval result.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Use a redacted packet when reproducing the task with business material. Our <a href=\"https:\/\/www.glbgpt.com\/hub\/ai-data-privacy\/\">Guide sur la protection des donn\u00e9es dans le domaine de l'IA<\/a> covers questions to ask about the platform and provider; accurate source handling alone does not establish privacy terms.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Test 3: plan recovery after an official page fails<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The third prompt simulated a browser agent facing HTTP 500, conflicting price claims and no authorized API key. It asked for exactly six steps, no more than 140 words and a concrete stopping condition. It explicitly requested planning rather than real tool execution.<\/p>\n\n\n\n<section id=\"t3-prompt\" data-text-card=\"copy\" style=\"border:1px solid #d4e1df;border-radius:17px;padding:20px;margin:22px 0;background:#132d32;color:#edf6f5;min-width:0\"><p style=\"margin:0 0 9px;font:700 12px\/1.5 system-ui;letter-spacing:.1em;text-transform:uppercase;color:#93cce2\">Entr\u00e9e exacte<\/p><div style=\"display:flex;flex-wrap:wrap;gap:10px;align-items:center;justify-content:space-between;margin-bottom:14px\"><strong style=\"font:700 19px\/1.4 Georgia,serif\">Recovery planning \u00b7 complete prompt<\/strong><button type=\"button\" data-copy onclick=\"var c=this.closest('[data-text-card]'),e=c.querySelector('[data-value]'),s=c.querySelector('[role=status]'),v=e.textContent;var fallback=function(){e.focus();var r=document.createRange();r.selectNodeContents(e);var sel=window.getSelection();sel.removeAllRanges();sel.addRange(r);s.textContent='Selected all. Press Ctrl\/Cmd+C.';};try{if(!navigator.clipboard||!navigator.clipboard.writeText){fallback();return;}Promise.race([navigator.clipboard.writeText(v),new Promise(function(_,reject){setTimeout(function(){reject(new Error('timeout'));},900);})]).then(function(){s.textContent='Copied.';},fallback);}catch(_){fallback();}\" style=\"border:1px solid #93b7aa;background:#d3e6cc;color:#183a32;border-radius:8px;padding:9px 15px;font:700 13px\/1.4 system-ui;cursor:pointer\">Copier l'invite<\/button><\/div><pre data-value tabindex=\"0\" contenteditable=\"true\" spellcheck=\"false\" role=\"textbox\" aria-multiline=\"true\" aria-label=\"Recovery planning \u00b7 complete prompt\" style=\"margin:0;padding:15px;border-radius:10px;border:1px solid #45636a;background:#0c2026;color:#edf6f5;font:13px\/1.75 ui-monospace,monospace;white-space:pre-wrap;overflow-wrap:anywhere;min-width:0\">Muse Wonder review test 3: recovery planning, not real tool execution.\nA browser agent must gather public product prices and prepare a comparison draft. The official pricing page returns HTTP 500. One marketplace lists input $4\/M and output $20\/M. A Reddit search snippet claims $1\/M. The user wants an evidence screenshot and a hands-on result. No authorized API key is available.\nWrite exactly six numbered steps, at most 140 words total. Include a bounded retry, retain the failed-page record, attribute marketplace pricing, do not treat the Reddit snippet as verified, and clearly distinguish preparation from a completed model test. Do not bypass login or fabricate screenshots\/outputs. End step 6 with a concrete stopping condition.<\/pre><span role=\"status\" aria-live=\"polite\" style=\"display:block;margin-top:10px;font:12px\/1.5 system-ui;color:#b9d3d1\">If automatic copy is blocked, select the text and press Ctrl\/Cmd+C.<\/span><\/section>\n\n\n\n<section id=\"t3-output\" data-text-card=\"result\" style=\"border:1px solid #d4e1df;border-radius:17px;padding:20px;margin:22px 0;background:#132d32;color:#edf6f5;min-width:0\"><p style=\"margin:0 0 9px;font:700 12px\/1.5 system-ui;letter-spacing:.1em;text-transform:uppercase;color:#93cce2\">R\u00e9sultats du mod\u00e8le<\/p><div style=\"display:flex;flex-wrap:wrap;gap:10px;align-items:center;justify-content:space-between;margin-bottom:14px\"><strong style=\"font:700 19px\/1.4 Georgia,serif\">Six-step response \u00b7 unchanged<\/strong><\/div><pre data-value tabindex=\"0\" style=\"margin:0;padding:15px;border-radius:10px;border:1px solid #45636a;background:#0c2026;color:#edf6f5;font:13px\/1.75 ui-monospace,monospace;white-space:pre-wrap;overflow-wrap:anywhere;min-width:0\">1. Retry official pricing page up to three times with brief delays; retain timestamp, URL, HTTP 500 status, and any genuine error-page screenshot as failure record.\n2. Record marketplace listing as marketplace-attributed pricing: input $4\/M and output $20\/M, with source URL\/date; do not present as official pricing.\n3. Log Reddit snippet claim of $1\/M as unverified lead only; exclude from verified comparison until corroborated by official or reputable source.\n4. Draft comparison table separating verified, attributed, and unverified entries, noting missing official confirmation.\n5. Prepare\u2014do not run\u2014model test plan, prompts, and scoring sheet; without authorized API key, do not bypass login or fabricate screenshots\/outputs; label status \u201cnot tested.\u201d\n6. Report draft as preparation only, not completed model test, and stop when three retries fail and no authorized key arrives by agreed deadline.<\/pre><\/section>\n\n\n\n<p class=\"wp-block-paragraph\">Muse Wonder produced six steps totaling 126 words, excluding step numbers. It bounded retries at three, retained the failed-page evidence, attributed marketplace prices, treated the Reddit snippet as unverified and kept preparation separate from a completed model test. All eight checks passed.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The plan\u2019s final deadline was not given a timestamp, so a real operator would still have to set it. More fundamentally, a good plan does not prove the agent can execute clicks, recover a tool connection or finish a long workflow. That is also why our <a href=\"https:\/\/www.glbgpt.com\/hub\/best-ai-assistant-tested-ranked\/\">Classement des assistants IA test\u00e9s<\/a> distinguishes different assistant tasks rather than treating every fluent answer as automation.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">What these runs did not measure<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">These examples cover three task types with no competing model. Code repair includes an explicit-requirements follow-up. The selected code, JSON and plan displayed \u201cTook 18s,\u201d \u201cTook 10s\u201d and \u201cTook 11s\u201d; those are platform labels, not independent latency measurements. We did not audit task-specific credit debits, provider token usage, persistent Active Memory, long context, multimodal files or live computer use.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The complete code, JSON and planning text is retained in the panels above. Use the exact inputs and verify the returned text when reproducing these examples; the accepted code is the follow-up version, and its result does not imply universal correctness for every application.<\/p>\n\n\n\n<h2 id=\"social-reaction\" class=\"wp-block-heading\">X benchmarks and Reddit reactions<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Le <a href=\"https:\/\/x.com\/AIatMeta\/status\/2109691148943999167\">AI at Meta launch thread on X<\/a> frames Wonder and Full around personal and professional agent work. That is the model maker\u2019s intended positioning. Our tests support some basic instruction-following observations, while the memory and execution claims remain separate.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Dans son <a href=\"https:\/\/x.com\/ArtificialAnlys\/status\/2109697083519447118\">original X assessment<\/a>, Artificial Analysis reports Intelligence Index v4.3 scores of 52 for Muse Wonder and 56 for Wonder Full at maximum reasoning effort, versus 48 for Muse Spark 1.3. The organization also distinguishes standard API access from Full in Muse Code. We read the original report; we did not reproduce its evaluations.<\/p>\n\n\n\n<figure style=\"margin:23px 0\"><a href=\"https:\/\/x.com\/ArtificialAnlys\/status\/2109697083519447118\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/static.futureshareai.com\/glb_features\/mcp\/4\/muse-wonder-x-artificial-analysis-english_bed712de88e6402886998d39b1deaaf4.webp\" alt=\"Original Artificial Analysis X post reporting Wonder 52, Wonder Full 56 and Spark 1.3 48, with author identity and the API versus Muse Code distinction.\" width=\"597\" height=\"213\" style=\"display:block;max-width:100%;height:auto;border:1px solid #d5dcd7;border-radius:12px\"><\/a><figcaption style=\"font-size:13px;line-height:1.6;color:#546a66;margin-top:9px\">Artificial Analysis\u2019s original X post reports 52 for Wonder and 56 for Full at maximum reasoning effort. These are third-party scores, not our test results. Captured October 14, 2026. Source: <a href=\"https:\/\/x.com\/ArtificialAnlys\/status\/2109697083519447118\">Original X post<\/a>.<\/figcaption><\/figure>\n\n\n\n<figure style=\"border:1px solid #d4e1df;border-radius:17px;padding:20px;margin:22px 0;background:#f0f5fb\"><figcaption style=\"font:700 19px\/1.4 Georgia,serif;color:#213d5a\">Artificial Analysis Intelligence Index v4.3<\/figcaption><p style=\"margin:7px 0 18px;font-size:13px;color:#516a82\">Reported scores \u00b7 maximum reasoning effort \u00b7 not our benchmark<\/p><div style=\"margin:15px 0\"><div style=\"display:flex;flex-wrap:wrap;gap:8px;justify-content:space-between;font-size:14px\"><strong>Muse Wonder Full \u00b7 56<\/strong><span>Highest of these three<\/span><\/div><div style=\"height:12px;background:#dae4ec;border-radius:10px;margin-top:7px\"><div style=\"height:100%;width:93.33333333333333%;background:#8b65bb;border-radius:10px\"><\/div><\/div><\/div><div style=\"margin:15px 0\"><div style=\"display:flex;flex-wrap:wrap;gap:8px;justify-content:space-between;font-size:14px\"><strong>Muse Wonder \u00b7 52<\/strong><span>4 points above Spark<\/span><\/div><div style=\"height:12px;background:#dae4ec;border-radius:10px;margin-top:7px\"><div style=\"height:100%;width:86.66666666666667%;background:#367fa8;border-radius:10px\"><\/div><\/div><\/div><div style=\"margin:15px 0\"><div style=\"display:flex;flex-wrap:wrap;gap:8px;justify-content:space-between;font-size:14px\"><strong>Muse Spark 1.3 \u00b7 48<\/strong><span>Earlier-family reference<\/span><\/div><div style=\"height:12px;background:#dae4ec;border-radius:10px;margin-top:7px\"><div style=\"height:100%;width:80%;background:#6b9584;border-radius:10px\"><\/div><\/div><\/div><p style=\"margin:14px 0 0;font-size:12px\">Bars use a display range of 0\u201360, not a percentage scale. Source: <a href=\"https:\/\/x.com\/ArtificialAnlys\/status\/2109697083519447118\">Artificial Analysis original post<\/a>.<\/p><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">The same report is positive about gains in agent tasks, but flags premium standard-model pricing and knowledge as a relative weakness. That combination is more informative than \u201cnew frontier model\u201d: a model can improve at agent work while still being the wrong economic choice for a frequent task. Its reported benchmark cost per task should not be reused as the price of our chat runs.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">On Reddit, the original LocalLLaMA post has been removed by moderators, but comments remain readable. <a href=\"https:\/\/www.reddit.com\/r\/LocalLLaMA\/comments\/1x4px4f\/comment\/pgfs3p3\/\">u\/noobiethe13 wrote<\/a>: \u201cwonder&#8217;s on the API, full is only through Muse Code subs for now. reviews should say which one they actually tried\u201d. The practical point matches Meta\u2019s access distinction: an unspecified \u201cWonder review\u201d can mix two different routes.<\/p>\n\n\n\n<figure style=\"margin:23px 0\"><a href=\"https:\/\/www.reddit.com\/r\/LocalLLaMA\/comments\/1x4px4f\/comment\/pgfs3p3\/\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/static.futureshareai.com\/glb_features\/mcp\/4\/muse-wonder-reddit-version-english_75859b996def404195e72ec7e54266a5.webp\" alt=\"Surviving Reddit comment by noobiethe13 asking reviewers to distinguish standard Wonder API access from Wonder Full in Muse Code.\" width=\"592\" height=\"117\" style=\"display:block;max-width:100%;height:auto;border:1px solid #d5dcd7;border-radius:12px\"><\/a><figcaption style=\"font-size:13px;line-height:1.6;color:#546a66;margin-top:9px\">u\/noobiethe13 asks reviewers to say which Wonder route they tested. The original post was removed; this comment remained readable. Captured October 14, 2026. Source: <a href=\"https:\/\/www.reddit.com\/r\/LocalLLaMA\/comments\/1x4px4f\/comment\/pgfs3p3\/\">Original Reddit comment<\/a>.<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Another commenter, <a href=\"https:\/\/www.reddit.com\/r\/LocalLLaMA\/comments\/1x4px4f\/comment\/pgfrjzb\/\">u\/ComplexType568<\/a>, wrote: \u201cHope they release a new open weights model and also an MoE model&#8230;\u201d That is an openness preference, not a measured quality result.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Moderator <a href=\"https:\/\/www.reddit.com\/r\/LocalLLaMA\/comments\/1x4px4f\/comment\/pgfy2yx\/\">u\/ttkciar<\/a> explained the removal: \u201cUntil such time that weights become available, this is off-topic for LocalLLaMA.\u201d These comments were posted October 13 and read on October 14. They explain this thread\u2019s concerns; they do not establish a Reddit-wide consensus or a reliability score.<\/p>\n\n\n\n<figure style=\"margin:23px 0\"><a href=\"https:\/\/www.reddit.com\/r\/LocalLLaMA\/comments\/1x4px4f\/comment\/pgfy2yx\/\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/static.futureshareai.com\/glb_features\/mcp\/4\/muse-wonder-reddit-moderator-english_f8913cda589948e09797847054b09d9b.webp\" alt=\"Reddit moderator ttkciar\u2019s original comment explaining that the thread is off-topic until model weights become available, shown in Reddit\u2019s English embed view.\" width=\"600\" height=\"168\" style=\"display:block;max-width:100%;height:auto;border:1px solid #d5dcd7;border-radius:12px\"><\/a><figcaption style=\"font-size:13px;line-height:1.6;color:#546a66;margin-top:9px\">Moderator u\/ttkciar explains the removal in the original comment, captured through Reddit\u2019s English embed view. This is one moderator\u2019s explanation, not community consensus. Captured October 14, 2026. Source: <a href=\"https:\/\/www.reddit.com\/r\/LocalLLaMA\/comments\/1x4px4f\/comment\/pgfy2yx\/\">Original Reddit comment<\/a>.<\/figcaption><\/figure>\n\n\n\n<h2 id=\"pricing\" class=\"wp-block-heading\">Muse Wonder pricing: tokens versus subscriptions<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Le <a href=\"https:\/\/openrouter.ai\/meta\/muse-wonder\">OpenRouter Muse Wonder listing<\/a> checked October 14 quotes $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. These are marketplace-listed API rates, attributed to OpenRouter rather than presented as a verified Meta invoice or checkout total.<\/p>\n\n\n\n<div style=\"display:grid;grid-template-columns:repeat(auto-fit,minmax(min(100%,210px),1fr));gap:12px;margin:20px 0\"><div style=\"border-top:4px solid #577aa4;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">API input<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">$4 \/ M<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">OpenRouter listing; uncached input tokens.<\/p><\/div><div style=\"border-top:4px solid #4b937a;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">Entr\u00e9e mise en cache<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">$0.20 \/ M<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">Listing rate; eligibility for caching matters.<\/p><\/div><div style=\"border-top:4px solid #8c75b6;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">API output<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">$20 \/ M<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">Listing rate; separate from subscription fees.<\/p><\/div><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">At those listed rates, an illustrative request with 10,000 uncached input tokens and 2,000 output tokens would cost <strong>$0.08<\/strong>: (10,000 \u00f7 1,000,000 \u00d7 $4) + (2,000 \u00f7 1,000,000 \u00d7 $20). This is arithmetic from displayed rates, excluding any other applicable charges. It is not the audited price of a completed review task.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Muse Code subscription access, standard API token billing and GlobalGPT credits are different units. Full\u2019s Muse Code access does not establish a per-token price for Full, and setting standard Wonder to high reasoning effort does not select Full. We did not verify a current Muse Code checkout total or a per-task GlobalGPT debit.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">If you run the same workflow daily, compare accepted output and retry cost as well as the headline rate. Our <a href=\"https:\/\/www.glbgpt.com\/hub\/which-ai-is-worth-paying-for\/\">Quelle IA vaut la peine d'\u00eatre pay\u00e9e ?<\/a> is useful for that broader subscription decision; a cheaper token rate can still be poor value if the task needs more attempts.<\/p>\n\n\n\n<h2 id=\"access\" class=\"wp-block-heading\">How to use Muse Wonder on GlobalGPT<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Le <a href=\"https:\/\/www.glbgpt.com\/home\/muse-wonder?inviter=hub_content_musewonder&amp;login=1\">Muse Wonder production entry on GlobalGPT<\/a> was published and enabled when checked. The original tasks and the code-repair follow-up completed there. Open it, confirm the Muse Wonder label and paste a complete prompt above. Judge the response against the stated checks before applying it to your own files.<\/p>\n\n\n\n<div style=\"border:1px solid #d4e1df;border-radius:17px;padding:20px;margin:22px 0;display:grid;grid-template-columns:repeat(auto-fit,minmax(min(100%,210px),1fr));gap:12px;background:#eef5f0\"><div style=\"min-width:0\"><span style=\"font:700 28px\/1.2 Georgia,serif;color:#578168\">01<\/span><strong style=\"display:block;margin:8px 0;font-size:16px\">Open Muse Wonder<\/strong><span style=\"font-size:14px\">Use the verified GlobalGPT entry.<\/span><\/div><div style=\"min-width:0\"><span style=\"font:700 28px\/1.2 Georgia,serif;color:#578168\">02<\/span><strong style=\"display:block;margin:8px 0;font-size:16px\">Run one task<\/strong><span style=\"font-size:14px\">Paste a complete prompt above.<\/span><\/div><div style=\"min-width:0\"><span style=\"font:700 28px\/1.2 Georgia,serif;color:#578168\">03<\/span><strong style=\"display:block;margin:8px 0;font-size:16px\">Verify the result<\/strong><span style=\"font-size:14px\">Execute checks or inspect source choices.<\/span><\/div><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">For code repair, spell out the allowed ID types and the expected equality rules. Start with a small sample, then check case handling, missing values, identity and input preservation. A specific acceptance list makes the returned function easier to verify and adapt.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">For direct API work, Meta\u2019s <a href=\"https:\/\/dev.meta.ai\/docs\/overview\">Model API overview<\/a> documents <code>https:\/\/api.meta.ai\/v1<\/code>, the standard model ID and a bearer key requirement. Our browser runs used the existing platform access; they do not establish that an arbitrary Meta API key or another gateway can call the same backend.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">GlobalGPT is convenient when you want to repeat a prompt and compare other models in one workspace. The <a href=\"https:\/\/www.glbgpt.com\/hub\/all-in-one-ai-models\/\">flux de travail des mod\u00e8les d'IA \u00ab tout-en-un \u00bb<\/a> describes that workflow. It does not make the chat interface equivalent to Muse Code\u2019s terminal agent or guarantee that every native memory\/tool feature is exposed.<\/p>\n\n\n\n<h2 id=\"verdict\" class=\"wp-block-heading\">Who should use it? Our verdict<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Try Muse Wonder for bounded coding, evidence cleanup and planning tasks.<\/strong> Our code run met the ordinary-ID requirements, the source run retained uncertainties and the planning run obeyed the recovery constraints. Those are concrete reasons to try it with a task you can verify.<\/p>\n\n\n\n<div style=\"display:grid;grid-template-columns:repeat(auto-fit,minmax(min(100%,210px),1fr));gap:12px;margin:20px 0\"><div style=\"border-top:4px solid #4b937a;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">Good first task<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">Code repair<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">A small function with explicit accepted inputs and checks.<\/p><\/div><div style=\"border-top:4px solid #577aa4;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">Useful research task<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">Source cleanup<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">Dated evidence, named sources and required unresolved fields.<\/p><\/div><div style=\"border-top:4px solid #c27b64;border-radius:12px;background:#f4f7f6;padding:17px;min-width:0\"><p style=\"margin:0 0 6px;font:700 12px\/1.4 system-ui;text-transform:uppercase;color:#465d61\">Needs a deeper test<\/p><strong style=\"font:700 25px\/1.3 Georgia,serif;color:#15343e\">Agent execution<\/strong><p style=\"margin:8px 0 0;font-size:14px;line-height:1.6\">Memory, browser recovery and long workflows remain untested.<\/p><\/div><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">Keep human review for production changes and define the accepted inputs before using generated code. Our follow-up checks support this specific implementation. If you need Wonder Full, evaluate the actual Muse Code route. For persistent memory or live computer use, run a separate end-to-end task and save its tool trace.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">For audio workflows, do not infer speech-generation quality from these text runs. A <a href=\"https:\/\/www.glbgpt.com\/hub\/ai-voice-agent-customer-support\/\">AI voice agents for customer support<\/a> workflow also needs speech recognition, a voice output route and its own latency checks; Wonder\u2019s documented audio input alone does not prove the whole stack.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Our early verdict is positive but specific: Muse Wonder handled the tested tasks well enough to earn a place in a practical trial. The next meaningful step is your own verified task, not a universal \u201cbest model\u201d claim from these examples.<\/p>\n\n\n\n<h2 id=\"faq\" class=\"wp-block-heading\">Muse Wonder review FAQ<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">What is Muse Wonder?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Muse Wonder is Meta\u2019s frontier agent model, launched October 12, 2026. Meta documents text, image, video, audio and PDF inputs, text output, and a 1,048,576-token context window.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Is Muse Wonder good for coding?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The code-repair follow-up passed 13 independent checks, including Symbol identity, object references, BigInt, order and input preservation. We made the identity requirements explicit and also executed all five supplied assertions successfully.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Did this review actually test Muse Wonder?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Yes. We tested code repair, conflicting-source JSON and recovery planning in GlobalGPT\u2019s production interface. The examples shown passed 29 independent checks. The code example is a follow-up with explicit identity requirements.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Can I use Muse Wonder on GlobalGPT?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Yes. We verified the published Muse Wonder entry and completed the original tasks and a code-repair follow-up there on October 14, 2026. The interface labels the model Muse Wonder; its exact provider backend version was not independently attested.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">What is the difference between Muse Wonder and Wonder Full?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Meta offers standard Wonder through Model API and Muse Code. The launch announcement limits Wonder Full to Muse Code subscribers. Our GlobalGPT runs do not establish that we tested Full.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">How much does Muse Wonder cost?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The OpenRouter listing checked October 14, 2026 quotes $4\/M input, $0.20\/M cached input and $20\/M output tokens. Those marketplace rates are separate from Muse Code subscriptions and GlobalGPT credits.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Does Muse Wonder generate images or video?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Meta\u2019s model documentation specifies multimodal inputs and text output. Understanding an image or video is different from generating one; this review tested text tasks only.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Did the tests prove Active Memory or computer use?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">No. The source packet was short, and the recovery task requested a plan only. We did not test cross-chat memory, a million-token input, multimodal files or live computer-use execution.<\/p>\n\n\n\n<script type=\"application\/ld+json\">{\n    \"@context\": \"https:\\\/\\\/schema.org\",\n    \"@type\": \"FAQPage\",\n    \"mainEntity\": [\n        {\n            \"@type\": \"Question\",\n            \"name\": \"What is Muse Wonder?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"Muse Wonder is Meta\\u2019s frontier agent model, launched October 12, 2026. Meta documents text, image, video, audio and PDF inputs, text output, and a 1,048,576-token context window.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Is Muse Wonder good for coding?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"The code-repair follow-up passed 13 independent checks, including Symbol identity, object references, BigInt, order and input preservation. We made the identity requirements explicit and also executed all five supplied assertions successfully.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Did this review actually test Muse Wonder?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"Yes. We tested code repair, conflicting-source JSON and recovery planning in GlobalGPT\\u2019s production interface. The examples shown passed 29 independent checks. The code example is a follow-up with explicit identity requirements.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Can I use Muse Wonder on GlobalGPT?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"Yes. We verified the published Muse Wonder entry and completed the original tasks and a code-repair follow-up there on October 14, 2026. The interface labels the model Muse Wonder; its exact provider backend version was not independently attested.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"What is the difference between Muse Wonder and Wonder Full?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"Meta offers standard Wonder through Model API and Muse Code. The launch announcement limits Wonder Full to Muse Code subscribers. Our GlobalGPT runs do not establish that we tested Full.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"How much does Muse Wonder cost?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"The OpenRouter listing checked October 14, 2026 quotes $4\\\/M input, $0.20\\\/M cached input and $20\\\/M output tokens. Those marketplace rates are separate from Muse Code subscriptions and GlobalGPT credits.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Does Muse Wonder generate images or video?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"Meta\\u2019s model documentation specifies multimodal inputs and text output. Understanding an image or video is different from generating one; this review tested text tasks only.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Did the tests prove Active Memory or computer use?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"No. The source packet was short, and the recovery task requested a plan only. We did not test cross-chat memory, a million-token input, multimodal files or live computer-use execution.\"\n            }\n        }\n    ]\n}<\/script>","protected":false},"excerpt":{"rendered":"<p>Muse Wonder is worth trying for code repair and source-based analysis. The examples below passed 29 independent checks: 13 for the code-repair follow-up and eight each for source reconciliation and recovery planning. With explicit identity requirements, the returned code preserved distinct Symbols and object references while keeping the requested order and format. Meta introduced Muse [&hellip;]<\/p>","protected":false},"author":16,"featured_media":20705,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_seopress_robots_primary_cat":"","_seopress_titles_title":"Muse Wonder Review: Real Tests, Pricing & Limits","_seopress_titles_desc":"Can Muse Wonder fix code and resolve conflicting sources? See our actual outputs, the Symbol bug and API costs. Choose a useful task before you switch.","_seopress_robots_index":"","footnotes":""},"categories":[7],"tags":[],"class_list":["post-20658","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-chat"],"acf":[],"_links":{"self":[{"href":"https:\/\/wp.glbgpt.com\/fr\/wp-json\/wp\/v2\/posts\/20658","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wp.glbgpt.com\/fr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wp.glbgpt.com\/fr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wp.glbgpt.com\/fr\/wp-json\/wp\/v2\/users\/16"}],"replies":[{"embeddable":true,"href":"https:\/\/wp.glbgpt.com\/fr\/wp-json\/wp\/v2\/comments?post=20658"}],"version-history":[{"count":6,"href":"https:\/\/wp.glbgpt.com\/fr\/wp-json\/wp\/v2\/posts\/20658\/revisions"}],"predecessor-version":[{"id":20708,"href":"https:\/\/wp.glbgpt.com\/fr\/wp-json\/wp\/v2\/posts\/20658\/revisions\/20708"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wp.glbgpt.com\/fr\/wp-json\/wp\/v2\/media\/20705"}],"wp:attachment":[{"href":"https:\/\/wp.glbgpt.com\/fr\/wp-json\/wp\/v2\/media?parent=20658"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wp.glbgpt.com\/fr\/wp-json\/wp\/v2\/categories?post=20658"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wp.glbgpt.com\/fr\/wp-json\/wp\/v2\/tags?post=20658"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}