{"id":20285,"date":"2026-10-02T03:14:30","date_gmt":"2026-10-02T07:14:30","guid":{"rendered":"https:\/\/wp.glbgpt.com\/?p=20285"},"modified":"2026-10-02T03:14:31","modified_gmt":"2026-10-02T07:14:31","slug":"claude-sonnet-5-5-vs-opus-5-5","status":"publish","type":"post","link":"https:\/\/wp.glbgpt.com\/pt-br\/hub\/claude-sonnet-5-5-vs-opus-5-5","title":{"rendered":"Claude Sonnet 5.5 x Opus 5.5: Pre\u00e7o, benchmarks e testes reais"},"content":{"rendered":"<style>.sonnet-opus{--ink:#172033;--muted:#64748b;--line:#dfe5ef;--navy:#111a36;--purple:#635bdb;--teal:#167d7b;max-width:1120px;margin:28px auto 72px;color:var(--ink);font:16px\/1.72 Inter,ui-sans-serif,-apple-system,BlinkMacSystemFont,\"Segoe UI\",Arial,sans-serif}.sonnet-opus h2{margin:54px 0 14px;padding-left:14px;border-left:5px solid var(--purple);font-size:30px;line-height:1.2}.sonnet-opus h3{line-height:1.25}.sonnet-opus p{margin:0 0 18px}.sonnet-opus a{color:#3557b7}.so-hero{padding:48px 54px 42px;border-radius:22px;background:linear-gradient(135deg,#111a36,#302d76 58%,#168482);color:#fff}.so-hero h1{margin:16px 0 12px;color:#fff;font-size:clamp(34px,5vw,56px);line-height:1.08}.so-hero p{color:#e8ebff;font-size:18px}.so-kicker{display:inline-block;padding:6px 10px;border:1px solid #ffffff4d;border-radius:999px;color:#e9eaff;font-size:12px;font-weight:800;letter-spacing:.08em;text-transform:uppercase}.so-quick,.so-method{margin:24px 0;padding:22px 25px;border:1px solid var(--line);border-radius:16px;background:#f4f2ff}.so-quick{border-left:6px solid var(--purple)}.so-quick h2{margin:0 0 10px;padding:0;border:0;font-size:24px}.so-quick li{margin:7px 0}.so-method{background:#eef6ff;border-color:#cbdcf8}.so-verdict{margin:24px 0;padding:24px 26px;border-radius:16px;background:var(--navy);color:#eff3ff}.so-verdict h2{margin:0 0 10px;padding:0;border:0;color:#fff;font-size:24px}.so-verdict p{color:#e8edff}.so-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:14px;margin:24px 0}.so-stat{padding:18px;border:1px solid var(--line);border-radius:14px;background:#fbfcff}.so-stat strong{display:block;font-size:28px;color:var(--purple)}.so-stat span{display:block;color:var(--muted);font-size:13px}.so-task{margin:28px 0;padding:22px;border:1px solid var(--line);border-radius:18px;background:#fff;box-shadow:0 10px 26px #10182812}.so-task-head{display:flex;justify-content:space-between;gap:16px;align-items:flex-start;margin-bottom:14px}.so-task h3{margin:3px 0;font-size:23px}.so-task-kicker{color:#5c55bd;font-size:11px;font-weight:800;letter-spacing:.08em;text-transform:uppercase}.so-pill{display:inline-flex;padding:5px 10px;border-radius:999px;background:#eef3ff;color:#334c9f;font-size:12px;font-weight:800}.so-observation{margin:14px 0;padding:12px 14px;border-left:4px solid var(--purple);background:#f5f3ff}.so-note{color:var(--muted);font-size:13px}.so-faq h3{margin-top:26px;font-size:20px}@media(max-width:780px){.so-hero{padding:34px 22px}.so-grid{grid-template-columns:1fr}.so-task-head{display:block}.so-pill{margin-top:10px}}<\/style>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><header class=\"so-hero\"><span class=\"so-kicker\">Compara\u00e7\u00e3o entre Claude e 5.5 \u00b7 tarefas de API correspondentes \u00b7 1\u00ba de outubro de 2026<\/span><h1>Claude Sonnet 5.5 x Opus 5.5: Pre\u00e7o, benchmarks e testes reais<\/h1><p>O Sonnet 5.5 e o Opus 5.5 pertencem \u00e0 gera\u00e7\u00e3o Claude 5.5, mas t\u00eam pre\u00e7os e posicionamentos diferentes. Esta compara\u00e7\u00e3o submete ambos aos mesmos cinco comandos e relata o que a rota realmente retornou.<\/p><\/header><\/div>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><nav aria-label=\"\u00cdndice\" style=\"margin:24px 0;padding:18px 22px;border:1px solid #dfe5ef;border-radius:14px;background:#fbfcff\"><strong>Nesta p\u00e1gina<\/strong><ul style=\"margin:10px 0 0;padding-left:20px\"><li><a href=\"#quick-answer\">Resposta r\u00e1pida<\/a><\/li><li><a href=\"#specs\">Pre\u00e7o oficial e especifica\u00e7\u00f5es do modelo<\/a><\/li><li><a href=\"#benchmarks\">Contexto oficial da avalia\u00e7\u00e3o comparativa<\/a><\/li><li><a href=\"#method\">M\u00e9todo da API com o mesmo prompt<\/a><\/li><li><a href=\"#aggregate\">Tempo, fichas e custo estimado<\/a><\/li><li><a href=\"#task-cards\">Cinco cart\u00f5es de tarefas<\/a><\/li><li><a href=\"#api-caveats\">Considera\u00e7\u00f5es sobre API e migra\u00e7\u00e3o<\/a><\/li><li><a href=\"#verdict\">O que as evid\u00eancias indicam<\/a><\/li><li><a href=\"#faq\">PERGUNTAS FREQUENTES<\/a><\/li><\/ul><\/nav><\/div>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>A diferen\u00e7a observada diz respeito, principalmente, ao custo e \u00e0 rapidez.<\/strong> Neste pacote de cinco prompts, o Sonnet 5.5 concluiu a tarefa mais rapidamente, utilizou menos tokens de sa\u00edda e apresentou a estimativa mais baixa de custo aritm\u00e9tico de solicita\u00e7\u00e3o. O Opus 5.5 utilizou mais tokens e tempo, enquanto suas respostas costumavam ser mais extensas. Os cart\u00f5es de tarefa mostram em quais casos esse texto adicional alterou o resultado \u00fatil e em quais ambos os modelos simplesmente foram aprovados.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">As instru\u00e7\u00f5es, o limite m\u00e1ximo de solicita\u00e7\u00f5es, a configura\u00e7\u00e3o de esfor\u00e7o, o ponto final e o projeto de uma execu\u00e7\u00e3o por modelo foram mantidos alinhados com o anterior <a href=\"https:\/\/www.glbgpt.com\/hub\/claude-opus-5-5-review\/\">An\u00e1lise do Claude Opus 5.5<\/a>.<\/p>\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=\"583\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/Claude-sonnet-5.5-1024x583.png\" class=\"wp-image-20335\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/Claude-sonnet-5.5-1024x583.png 1024w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/Claude-sonnet-5.5-300x171.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/Claude-sonnet-5.5-768x437.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/Claude-sonnet-5.5-18x10.png 18w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/Claude-sonnet-5.5-1536x874.png 1536w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/Claude-sonnet-5.5.jpg 1699w\" 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-white-color has-text-color has-background has-link-color wp-element-button\" href=\"https:\/\/www.glbgpt.com\/home\/claude-sonnet-5-5?inviter=hub_sonnet55&amp;login=1\" style=\"background:linear-gradient(135deg,rgb(255,203,112) 0%,rgb(199,81,192) 78%,rgb(65,88,208) 100%)\"><strong>Experimente o Claude Sonnet 5.5 agora no GlobalGPT<\/strong><\/a><\/div>\n<\/div>\n<\/div>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><section class=\"so-quick\" id=\"quick-answer\"><h2>Resposta r\u00e1pida<\/h2><ul><li><strong>Velocidade nesta corrida em dupla:<\/strong> O Sonnet 5.5 concluiu cinco solicita\u00e7\u00f5es consecutivas em 31,930 segundos localmente; o Opus 5.5 levou 47,725 segundos.<\/li><li><strong>Tokens de sa\u00edda relatados pela rota:<\/strong> O Sonnet 5.5 utilizou 2.686; o Opus 5.5 utilizou 3.952. Os campos de reflex\u00e3o foram 987 e 2.214, respectivamente.<\/li><li><strong>Estimativa aritm\u00e9tica do pre\u00e7o de tabela:<\/strong> os cinco pedidos do Sonnet totalizaram cerca de $0,02826; os cinco pedidos do Opus totalizaram cerca de $0,08184, utilizando as tarifas oficiais padr\u00e3o por token e excluindo cache, cr\u00e9ditos, impostos e margem de lucro.<\/li><li><strong>Resultado da tarefa:<\/strong> Ambos os modelos foram aprovados nas tarefas de extra\u00e7\u00e3o, JSON e matem\u00e1tica. O Sonnet manteve o formato solicitado de dois par\u00e1grafos em chin\u00eas de maneira mais clara; os resultados de programa\u00e7\u00e3o eram ambos utiliz\u00e1veis, mas diferiam em profundidade e na explica\u00e7\u00e3o de casos extremos.<\/li><li><strong>Migra\u00e7\u00e3o da API:<\/strong> O racioc\u00ednio n\u00e3o pode ser desativado; a escolha for\u00e7ada de ferramentas \u00e9 rejeitada, e os or\u00e7amentos de fichas incluem o racioc\u00ednio. Verifique o <a href=\"#api-caveats\">considera\u00e7\u00f5es sobre a migra\u00e7\u00e3o<\/a> antes de mudar.<\/li><li><strong>Limite de decis\u00e3o:<\/strong> Trata-se de uma execu\u00e7\u00e3o por tarefa por meio de uma rota de terceiros. Ela mede as respostas observadas ao prompt, o tempo real local e o uso relatado pela rota \u2014 n\u00e3o se trata de uma pontua\u00e7\u00e3o universal de capacidade.<\/li><\/ul><\/section><\/div>\n\n\n\n<h2 id=\"specs\" class=\"wp-block-heading\">Pre\u00e7o oficial e especifica\u00e7\u00f5es do modelo<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">As fichas t\u00e9cnicas atuais do Anthropic listam ambos os modelos com uma janela de contexto de 1 milh\u00e3o de tokens e uma sa\u00edda m\u00e1xima de 128 mil. O Sonnet 5.5 \u00e9 identificado como <strong>R\u00e1pido<\/strong> com alto n\u00edvel de esfor\u00e7o padr\u00e3o; o Opus 5.5 \u00e9 identificado como <strong>Moderado<\/strong> com um esfor\u00e7o padr\u00e3o m\u00e9dio. As taxas padr\u00e3o dos tokens fazem com que o Sonnet 5.5 tenha metade do pre\u00e7o de entrada e de sa\u00edda do Opus 5.5.<\/p>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><div style=\"overflow-x:auto;margin:20px 0\"><table style=\"width:100%;border-collapse:collapse;min-width:680px\"><thead><tr><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Modelo<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">ID da API<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Indica\u00e7\u00e3o oficial de lat\u00eancia<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Entrada \/ sa\u00edda<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Contexto \/ pot\u00eancia m\u00e1xima<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Esfor\u00e7o padr\u00e3o<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Leitura do cache<\/th><\/tr><\/thead><tbody><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Sonnet 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><code>claude-soneto-5-5<\/code><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Ativo \u00b7 R\u00e1pido<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">$2 \/ $10 por MTok<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">1M \/ 128K<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Alto<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">$0.20 \/ MTok<\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Opus 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><code>claude-opus-5-5<\/code><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Ativo \u00b7 Moderado<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">$4 \/ $20 por MTok<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">1M \/ 128K<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">M\u00e9dio<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">$0.20 \/ MTok<\/td><\/tr><\/tbody><\/table><\/div><\/div>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" width=\"922\" height=\"244\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/official-sonnet-55-price-specs.png\" alt=\"Documenta\u00e7\u00e3o oficial do Claude Sonnet 5.5: contexto de 1M, sa\u00edda de 128K, entrada de $2 e sa\u00edda de $10 por milh\u00e3o de tokens\" class=\"wp-image-20318\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/official-sonnet-55-price-specs.png 922w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/official-sonnet-55-price-specs-300x79.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/official-sonnet-55-price-specs-768x203.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/official-sonnet-55-price-specs-18x5.png 18w\" sizes=\"(max-width: 922px) 100vw, 922px\" \/><figcaption class=\"wp-element-caption\"><a href=\"https:\/\/platform.claude.com\/docs\/en\/models\/sonnet-5-5\/overview\">Ficha t\u00e9cnica do modelo Sonnet 5.5 da Anthropic<\/a>. Pre\u00e7os padr\u00e3o da API por milh\u00e3o de tokens; dados coletados em 2 de outubro de 2026.<\/figcaption><\/figure>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" width=\"922\" height=\"244\" src=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/official-opus-55-price-specs.png\" alt=\"Documenta\u00e7\u00e3o oficial do Claude Opus 5.5: contexto de 1M, sa\u00edda de 128K, entrada de $4 e sa\u00edda de $20 por milh\u00e3o de tokens\" class=\"wp-image-20319\" srcset=\"https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/official-opus-55-price-specs.png 922w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/official-opus-55-price-specs-300x79.png 300w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/official-opus-55-price-specs-768x203.png 768w, https:\/\/wp.glbgpt.com\/wp-content\/uploads\/2026\/10\/official-opus-55-price-specs-18x5.png 18w\" sizes=\"(max-width: 922px) 100vw, 922px\" \/><figcaption class=\"wp-element-caption\"><a href=\"https:\/\/platform.claude.com\/docs\/en\/models\/opus-5-5\/overview\">Placa modelo Opus 5.5 da Anthropic<\/a>. Pre\u00e7os padr\u00e3o da API por milh\u00e3o de tokens; dados coletados em 2 de outubro de 2026.<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Para um exemplo simples, sem cache, com 100.000 tokens de entrada e 20.000 tokens de sa\u00edda, as taxas de tokens listadas correspondem a cerca de $0,40 no Sonnet 5,5 e $0,80 no Opus 5,5. Consulte o <a href=\"https:\/\/www.glbgpt.com\/hub\/claude-ai-pricing-2026-the-ultimate-guide-to-plans-api-costs-and-limits\/\">Guia de pre\u00e7os e limites do Claude<\/a> para o contexto do plano.<\/p>\n\n\n\n<h2 id=\"benchmarks\" class=\"wp-block-heading\">Contexto oficial da avalia\u00e7\u00e3o comparativa<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Anthropic\u2019s <a href=\"https:\/\/www.anthropic.com\/claude-sonnet-5-5\">An\u00fancio Sonnet 5.5<\/a> coloca os dois modelos em uma \u00fanica tabela fornecida pelo prestador. O padr\u00e3o das linhas varia de acordo com o par\u00e2metro de refer\u00eancia e a configura\u00e7\u00e3o de esfor\u00e7o.<\/p>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><div style=\"overflow-x:auto;margin:20px 0\"><table style=\"width:100%;border-collapse:collapse;min-width:680px\"><thead><tr><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">\u00cdndice de refer\u00eancia divulgado pela Anthropic<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Sonnet 5.5<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Opus 5.5<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Medida<\/th><\/tr><\/thead><tbody><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Terminal-Bench 4.0<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">70.6%<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">66,4%\u00b9<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Codifica\u00e7\u00e3o de terminal ag\u00eanica<\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">FrontierCode v1.1 (Principal)<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">46,21 TP40T Max\u00b2 \/ 52,11 TP40T Xhigh<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">54.4%<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Se as altera\u00e7\u00f5es no c\u00f3digo seriam incorporadas<\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">CursorBench 4.0<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">55.5%<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">57.8%<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Tarefas amb\u00edguas de codifica\u00e7\u00e3o de v\u00e1rios arquivos<\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">GDPval-AA v2.1<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">1844<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">1846<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Trabalho do conhecimento em diversas profiss\u00f5es<\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">O \u00faltimo exame da humanidade<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">64,51 TP40T com ferramentas<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">67,71 TP40T com ferramentas<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Racioc\u00ednio multidisciplinar<\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">OSWorld 2.1<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">80.1% parcial<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">81,81 TP40T parcial<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Tarefas que envolvem o uso do computador<\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Cartografia<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">61,61 TP40T sem ferramentas<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">64,41 TP40T sem ferramentas<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Reconhecimento visual de gr\u00e1ficos<\/td><\/tr><\/tbody><\/table><\/div><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">\u00b9 O Opus 5.5 Terminal-Bench utiliza o n\u00edvel de esfor\u00e7o \u201cxhigh\u201d; \u00b2 O Sonnet 5.5 FrontierCode apresenta 46,2% no n\u00edvel \u201cMax\u201d e 52,1% no n\u00edvel de esfor\u00e7o \u201cXhigh\u201d. Os valores do OSWorld est\u00e3o marcados como parciais. Trata-se de um contexto espec\u00edfico do provedor, e n\u00e3o de uma repeti\u00e7\u00e3o independente com esfor\u00e7o equivalente.<\/p>\n\n\n\n<h2 id=\"method\" class=\"wp-block-heading\">M\u00e9todo da API com o mesmo prompt<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Cada modelo recebeu as mesmas cinco instru\u00e7\u00f5es por meio de <code>POST https:\/\/anywhere.broly.ai\/v1\/messages<\/code>. O cliente enviou uma mensagem ao usu\u00e1rio, <code>max_tokens: 8192<\/code>, <code>output_config.effort: alto<\/code>, e sem ferramentas. Todas as solicita\u00e7\u00f5es retornaram o c\u00f3digo HTTP 200 e <code>fim_do_turno<\/code>.<\/p>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><div class=\"so-method\"><strong>Limite de medi\u00e7\u00e3o:<\/strong> O tempo decorrido \u00e9 o tempo real local de ponta a ponta, e n\u00e3o a lat\u00eancia do provedor. As contagens de tokens e o \u201ccampo de reflex\u00e3o\u201d representam o uso relatado pela rota.<\/div><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">A abordagem de testes segue o conceito mais amplo de <a href=\"https:\/\/www.glbgpt.com\/hub\/how-globalgpt-tests-ai-models\/\">Fluxo de trabalho de teste do modelo GlobalGPT<\/a>. Um outro <a href=\"https:\/\/www.glbgpt.com\/hub\/claude-api-guide\/\">Guia da API Claude<\/a> abrange a configura\u00e7\u00e3o de solicita\u00e7\u00f5es e a contabilidade de tokens.<\/p>\n\n\n\n<h2 id=\"aggregate\" class=\"wp-block-heading\">Tempo, fichas e custo estimado<\/h2>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><div style=\"overflow-x:auto;margin:20px 0\"><table style=\"width:100%;border-collapse:collapse;min-width:680px\"><thead><tr><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Rota<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Hora local das cinco<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Tokens de entrada<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Tokens de sa\u00edda<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Pensamento relatado<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Custo estimado da solicita\u00e7\u00e3o*<\/th><\/tr><\/thead><tbody><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Sonnet 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">31,930 s<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">700<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">2,686<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">987<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>$0.02826<\/strong><\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Opus 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">47,725 s<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">700<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">3,952<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">2,214<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>$0.08184<\/strong><\/td><\/tr><\/tbody><\/table><\/div><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">* Estimativa com base nos tokens relatados pela rota e nas taxas padr\u00e3o oficiais. O Sonnet 5.5 apresentou uma redu\u00e7\u00e3o de 33,1% no tempo decorrido local e de 32,0% na contagem de tokens de sa\u00edda neste grupo.<\/p>\n\n\n\n<h2 id=\"task-cards\" class=\"wp-block-heading\">Cinco cart\u00f5es de tarefas correspondentes<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Cada cart\u00e3o re\u00fane a tarefa, a observa\u00e7\u00e3o, a hora local, os tokens de rota registrados, o status e os limites.<\/p>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><section class=\"so-task\" id=\"coding-debug\"><div class=\"so-task-head\"><div><span class=\"so-task-kicker\">Tarefa 01 \u00b7 Depura\u00e7\u00e3o em Python<\/span><h3>Os modelos s\u00e3o capazes de corrigir uma fun\u00e7\u00e3o de deduplica\u00e7\u00e3o de tipo misto?<\/h3><\/div><span class=\"so-pill\">Prompt correspondente \u00b7 uma execu\u00e7\u00e3o de cada<\/span><\/div><p><strong>Configura\u00e7\u00e3o da tarefa:<\/strong> A instru\u00e7\u00e3o exata foi copiada da vers\u00e3o anterior <a href=\"https:\/\/www.glbgpt.com\/hub\/claude-opus-5-5-review\/\">An\u00e1lise do Claude Opus 5.5<\/a> pacote de teste.<\/p><div style=\"overflow-x:auto;margin:20px 0\"><table style=\"width:100%;border-collapse:collapse;min-width:680px\"><thead><tr><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Modelo<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Tempo \/ uso da rota<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">R\u00f3tulo de resultado curto<\/th><\/tr><\/thead><tbody><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Sonnet 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">8,679 s<br>145 de entrada \/ 830 de sa\u00edda<br>0 reflex\u00e3o<br>HTTP 200 \u00b7 fim_do_turno<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Fun\u00e7\u00e3o corrigida, al\u00e9m de dois exerc\u00edcios; explica\u00e7\u00e3o mais sucinta.<\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Opus 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">15,844 s<br>145 de entrada \/ 1.478 de sa\u00edda<br>700 reflex\u00f5es<br>HTTP 200 \u00b7 fim_do_turno<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Fun\u00e7\u00e3o corrigida, al\u00e9m de dois testes; mais discuss\u00e3o sobre casos extremos.<\/td><\/tr><\/tbody><\/table><\/div><p class=\"so-observation\"><strong>Compara\u00e7\u00e3o observada:<\/strong> Ambos retornaram uma fun\u00e7\u00e3o corrigida e dois testes. O Sonnet 5.5 utilizou chaves com marca\u00e7\u00e3o de tipo, <code>casefold()<\/code>, e uma alternativa em forma de lista em uma resposta mais curta; o Opus 5.5 utilizou mais tokens de sa\u00edda para explicar casos extremos adicionais. Nenhuma das duas execu\u00e7\u00f5es indica um vencedor geral em termos de codifica\u00e7\u00e3o.<\/p><p class=\"so-note\"><strong>Limite:<\/strong> Uma \u00fanica pequena corre\u00e7\u00e3o em Python verifica casos extremos espec\u00edficos, e n\u00e3o a confiabilidade entre reposit\u00f3rios ou na codifica\u00e7\u00e3o orientada por ferramentas.<\/p><\/section><\/div>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><section class=\"so-task\" id=\"long-context-extract\"><div class=\"so-task-head\"><div><span class=\"so-task-kicker\">Tarefa 02 \u00b7 Extra\u00e7\u00e3o com aterramento<\/span><h3>Os modelos conseguem preservar os cinco fatos fornecidos sem acrescentar afirma\u00e7\u00f5es?<\/h3><\/div><span class=\"so-pill\">Prompt correspondente \u00b7 uma execu\u00e7\u00e3o de cada<\/span><\/div><p><strong>Configura\u00e7\u00e3o da tarefa:<\/strong> A instru\u00e7\u00e3o exata foi copiada da vers\u00e3o anterior <a href=\"https:\/\/www.glbgpt.com\/hub\/claude-opus-5-5-review\/\">An\u00e1lise do Claude Opus 5.5<\/a> pacote de teste.<\/p><div style=\"overflow-x:auto;margin:20px 0\"><table style=\"width:100%;border-collapse:collapse;min-width:680px\"><thead><tr><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Modelo<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Tempo \/ uso da rota<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">R\u00f3tulo de resultado curto<\/th><\/tr><\/thead><tbody><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Sonnet 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">3,569 s<br>210 de entrada \/ 209 de sa\u00edda<br>0 reflex\u00e3o<br>HTTP 200 \u00b7 fim_do_turno<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Cinco marcadores numerados; os fatos apresentados foram mantidos.<\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Opus 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">4,374 s<br>210 de entrada \/ 312 de sa\u00edda<br>101 formas de pensar<br>HTTP 200 \u00b7 fim_do_turno<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Cinco marcadores numerados; os fatos apresentados foram mantidos.<\/td><\/tr><\/tbody><\/table><\/div><p class=\"so-observation\"><strong>Compara\u00e7\u00e3o observada:<\/strong> Ambos geraram exatamente cinco marcadores numerados e se mantiveram dentro dos fatos fornecidos. O Sonnet 5.5 utilizou 209 tokens de sa\u00edda, contra os 312 do Opus 5.5, mas o prompt era uma nota curta, e n\u00e3o uma entrada com contexto amplo.<\/p><p class=\"so-note\"><strong>Limite:<\/strong> Trata-se de uma verifica\u00e7\u00e3o fundamentada de extra\u00e7\u00e3o e formata\u00e7\u00e3o; n\u00e3o constitui evid\u00eancia de desempenho em contextos com milh\u00f5es de tokens.<\/p><\/section><\/div>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><section class=\"so-task\" id=\"structured-json\"><div class=\"so-task-head\"><div><span class=\"so-task-kicker\">Tarefa 03 \u00b7 Conformidade com JSON<\/span><h3>Os modelos conseguem retornar exatamente a estrutura solicitada?<\/h3><\/div><span class=\"so-pill\">Prompt correspondente \u00b7 uma execu\u00e7\u00e3o de cada<\/span><\/div><p><strong>Configura\u00e7\u00e3o da tarefa:<\/strong> A instru\u00e7\u00e3o exata foi copiada da vers\u00e3o anterior <a href=\"https:\/\/www.glbgpt.com\/hub\/claude-opus-5-5-review\/\">An\u00e1lise do Claude Opus 5.5<\/a> pacote de teste.<\/p><div style=\"overflow-x:auto;margin:20px 0\"><table style=\"width:100%;border-collapse:collapse;min-width:680px\"><thead><tr><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Modelo<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Tempo \/ uso da rota<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">R\u00f3tulo de resultado curto<\/th><\/tr><\/thead><tbody><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Sonnet 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">5,052 s<br>128 de entrada \/ 260 de sa\u00edda<br>0 reflex\u00e3o<br>HTTP 200 \u00b7 fim_do_turno<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">JSON analis\u00e1vel; chaves solicitadas e contagem de itens.<\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Opus 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">8,276 s<br>128 de entrada \/ 622 de sa\u00edda<br>390 reflex\u00f5es<br>HTTP 200 \u00b7 fim_do_turno<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">JSON analis\u00e1vel; chaves solicitadas e contagem de itens.<\/td><\/tr><\/tbody><\/table><\/div><p class=\"so-observation\"><strong>Compara\u00e7\u00e3o observada:<\/strong> Ambos retornaram um JSON analis\u00e1vel com as chaves solicitadas e duas vantagens e duas desvantagens. O Sonnet 5.5 foi mais conciso, com 260 tokens de sa\u00edda; as sequ\u00eancias de caracteres descrevem uma situa\u00e7\u00e3o fict\u00edcia de avalia\u00e7\u00e3o e n\u00e3o s\u00e3o informa\u00e7\u00f5es sobre o produto.<\/p><p class=\"so-note\"><strong>Limite:<\/strong> A execu\u00e7\u00e3o desse esquema uma \u00fanica vez n\u00e3o permite avaliar a confiabilidade da sa\u00edda estruturada em chamadas de ferramentas ou conversas longas.<\/p><\/section><\/div>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><section class=\"so-task\" id=\"math-reasoning\"><div class=\"so-task-head\"><div><span class=\"so-task-kicker\">Tarefa 04 \u00b7 Aritm\u00e9tica<\/span><h3>Os modelos conseguem aplicar a sequ\u00eancia arredondada de lotes at\u00e9 a resposta final?<\/h3><\/div><span class=\"so-pill\">Prompt correspondente \u00b7 uma execu\u00e7\u00e3o de cada<\/span><\/div><p><strong>Configura\u00e7\u00e3o da tarefa:<\/strong> A instru\u00e7\u00e3o exata foi copiada da vers\u00e3o anterior <a href=\"https:\/\/www.glbgpt.com\/hub\/claude-opus-5-5-review\/\">An\u00e1lise do Claude Opus 5.5<\/a> pacote de teste.<\/p><div style=\"overflow-x:auto;margin:20px 0\"><table style=\"width:100%;border-collapse:collapse;min-width:680px\"><thead><tr><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Modelo<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Tempo \/ uso da rota<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">R\u00f3tulo de resultado curto<\/th><\/tr><\/thead><tbody><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Sonnet 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">2,947 s<br>89 de entrada \/ 163 de sa\u00edda<br>0 reflex\u00e3o<br>HTTP 200 \u00b7 fim_do_turno<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Resultado correto: 67; a resposta final deve constar em uma linha separada.<\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Opus 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">3,830 s<br>89 de entrada \/ 257 de sa\u00edda<br>74 reflex\u00f5es<br>HTTP 200 \u00b7 fim_do_turno<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Resultado correto: 67; a resposta final deve constar em uma linha separada.<\/td><\/tr><\/tbody><\/table><\/div><p class=\"so-observation\"><strong>Compara\u00e7\u00e3o observada:<\/strong> Ambos calcularam 67 e colocaram a resposta final em uma linha separada. O Sonnet 5.5 utilizou 163 tokens de sa\u00edda, contra os 257 do Opus 5.5, sem que se observasse diferen\u00e7a na precis\u00e3o nesse prompt.<\/p><p class=\"so-note\"><strong>Limite:<\/strong> Uma sequ\u00eancia aritm\u00e9tica por si s\u00f3 n\u00e3o \u00e9 suficiente para avaliar a confiabilidade do racioc\u00ednio geral.<\/p><\/section><\/div>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><section class=\"so-task\" id=\"chinese-seo-outline\"><div class=\"so-task-head\"><div><span class=\"so-task-kicker\">Tarefa 05 \u00b7 Seguir o formato chin\u00eas<\/span><h3>Ser\u00e1 que o roteiro pode manter a estrutura solicitada de dois par\u00e1grafos?<\/h3><\/div><span class=\"so-pill\">Prompt correspondente \u00b7 uma execu\u00e7\u00e3o de cada<\/span><\/div><p><strong>Configura\u00e7\u00e3o da tarefa:<\/strong> A instru\u00e7\u00e3o exata foi copiada da vers\u00e3o anterior <a href=\"https:\/\/www.glbgpt.com\/hub\/claude-opus-5-5-review\/\">An\u00e1lise do Claude Opus 5.5<\/a> pacote de teste.<\/p><div style=\"overflow-x:auto;margin:20px 0\"><table style=\"width:100%;border-collapse:collapse;min-width:680px\"><thead><tr><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Modelo<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">Tempo \/ uso da rota<\/th><th style=\"background:#f1f3fa;border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top;font-size:13px\">R\u00f3tulo de resultado curto<\/th><\/tr><\/thead><tbody><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Sonnet 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">11,683 s<br>128 entradas \/ 1.224 sa\u00eddas<br>987 reflex\u00f5es<br>HTTP 200 \u00b7 fim_do_turno<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Dois par\u00e1grafos em chin\u00eas; sem formata\u00e7\u00e3o adicional.<\/td><\/tr><tr><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\"><strong>Claude Opus 5,5<\/strong><\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">15,401 s<br>128 entradas \/ 1.283 sa\u00eddas<br>949 reflex\u00f5es<br>HTTP 200 \u00b7 fim_do_turno<\/td><td style=\"border:1px solid #dfe5ef;padding:11px 12px;text-align:left;vertical-align:top\">Abordei os pontos; adicionei Markdown e uma nota em ingl\u00eas.<\/td><\/tr><\/tbody><\/table><\/div><p class=\"so-observation\"><strong>Compara\u00e7\u00e3o observada:<\/strong> O Sonnet 5.5 entregou os dois par\u00e1grafos em chin\u00eas solicitados, sem formata\u00e7\u00e3o adicional. O Opus 5.5 tamb\u00e9m abordou os pontos solicitados, mas acrescentou formata\u00e7\u00e3o em Markdown e uma nota em ingl\u00eas. Este registro reflete o formato seguido em uma \u00fanica execu\u00e7\u00e3o, e n\u00e3o a qualidade geral do texto em chin\u00eas.<\/p><p class=\"so-note\"><strong>Limite:<\/strong> A instru\u00e7\u00e3o pede uma introdu\u00e7\u00e3o sobre o Opus 5.5; portanto, o texto em si n\u00e3o \u00e9 um par\u00e2metro de refer\u00eancia de linguagem neutra.<\/p><\/section><\/div>\n\n\n\n<h2 id=\"api-caveats\" class=\"wp-block-heading\">Considera\u00e7\u00f5es sobre API e migra\u00e7\u00e3o<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">O Sonnet 5.5 executa o pensamento adaptativo por padr\u00e3o; a op\u00e7\u00e3o \u201cthinking: disabled\u201d retorna um erro 400, e o par\u00e2metro `max_tokens` abrange tanto o pensamento quanto o texto da resposta. A escolha for\u00e7ada de ferramenta (`any\/tool`) \u00e9 rejeitada. Analise os blocos de conte\u00fado por tipo e redefina os limites de tokens antes de alternar.<\/p>\n\n\n\n<h2 id=\"verdict\" class=\"wp-block-heading\">O que as evid\u00eancias indicam<\/h2>\n\n\n\n<div class=\"sonnet-opus\" style=\"margin:0;max-width:none\"><div class=\"so-verdict\"><h2>Veredicto no n\u00edvel da tarefa<\/h2><p><strong>Sonnet 5.5:<\/strong> custo e tempo medidos mais baixos nessa rota, com conformidade perfeita com o formato na tarefa chinesa.<\/p><p><strong>Opus 5.5:<\/strong> mais caro e mais abrangente aqui; as s\u00e9ries de testes de desempenho oficiais continuam \u00e0 frente em v\u00e1rias tarefas de dura\u00e7\u00e3o indeterminada.<\/p><p>Use as evid\u00eancias no n\u00edvel da tarefa para escolher um ponto de partida e, em seguida,avalie a carga de trabalho que \u00e9 importante para voc\u00ea.<\/p><\/div><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">Para o contexto adjacente, consulte o <a href=\"https:\/\/www.glbgpt.com\/hub\/claude-opus-5-vs-fable-5-vs-sonnet-5\/\">Compara\u00e7\u00e3o da fam\u00edlia Claude<\/a>, <a href=\"https:\/\/www.glbgpt.com\/hub\/claude-opus-5-review\/\">Resenha de Opus 5<\/a>, e <a href=\"https:\/\/www.glbgpt.com\/hub\/claude-fable-5-1\/\">An\u00e1lise do Fable 5.1<\/a>.<\/p>\n\n\n\n<h2 id=\"faq\" class=\"wp-block-heading\">PERGUNTAS FREQUENTES<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">Qual modelo foi mais r\u00e1pido no teste comparativo?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">O Sonnet 5.5 apresentou o menor tempo total local: 31,930 segundos contra 47,725 segundos do Opus 5.5 em cinco solicita\u00e7\u00f5es consecutivas. Isso inclui a rota e o caminho de rede utilizados neste caso, portanto, n\u00e3o se trata de lat\u00eancia do lado do provedor.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Qual modelo utilizou menos tokens de sa\u00edda?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">O Sonnet 5.5 utilizou 2.686 tokens de sa\u00edda relatados pela rota nas cinco tarefas, em compara\u00e7\u00e3o com 3.952 do Opus 5.5. A contagem de tokens por si s\u00f3 n\u00e3o comprova a qualidade da resposta.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Qual modelo ficou mais barato nessa s\u00e9rie de testes?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Utilizando as taxas oficiais padr\u00e3o da API e as contagens de entrada\/sa\u00edda retornadas, as cinco solicita\u00e7\u00f5es Sonnet 5.5 totalizam $0,02826, contra $0,08184 para o Opus 5.5. O c\u00e1lculo exclui encargos de cache, cr\u00e9ditos de plataforma, impostos e margem de lucro.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">O Sonnet 5.5 supera o Opus 5.5 em todos os testes de desempenho?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">A tabela do Anthropic apresenta resultados variados de acordo com o benchmark e a configura\u00e7\u00e3o de esfor\u00e7o: O Sonnet 5.5 apresenta desempenho superior no Terminal-Bench 4.0, enquanto o Opus 5.5 apresenta desempenho superior no FrontierCode, CursorBench, GDPval-AA, Humanity\u2019s Last Exam, OSWorld e Chartography. Esses s\u00e3o resultados informados pelos fornecedores, n\u00e3o uma repeti\u00e7\u00e3o independente dos testes.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">A Sonnet 5.5 \u00e9 um substituto direto da API?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">N\u00e3o. O guia de migra\u00e7\u00e3o indica que o \u201cthinking\u201d \u00e9 executado por padr\u00e3o; a op\u00e7\u00e3o \u201cthinking: disabled\u201d retorna um erro 400; a escolha for\u00e7ada de ferramenta \u201cany\/tool\u201d \u00e9 rejeitada; e os clientes devem analisar os blocos de conte\u00fado por tipo. Reajuste os or\u00e7amentos de tokens e os ciclos das ferramentas antes de fazer a mudan\u00e7a.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Esse foi um teste de desempenho em contexto extenso?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">N\u00e3o. A solicita\u00e7\u00e3o de extra\u00e7\u00e3o cont\u00e9m um pequeno trecho. Ela verifica a fundamenta\u00e7\u00e3o factual e a formata\u00e7\u00e3o exata; n\u00e3o avalia o comportamento pr\u00f3ximo \u00e0 janela de contexto documentada de 1 milh\u00e3o de tokens.<\/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\": \"Which model was faster in the matched test?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"Sonnet 5.5 had the lower local total: 31.930 seconds versus 47.725 seconds for Opus 5.5 across five sequential requests. This includes the route and network path used here, so it is not provider-side latency.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Which model used fewer output tokens?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"Sonnet 5.5 used 2,686 route-reported output tokens across the five tasks, compared with 3,952 for Opus 5.5. Token count alone does not prove answer quality.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Which model was cheaper in this run?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"Using official standard API rates and the returned input\\\/output counts, the five Sonnet 5.5 requests estimate to $0.02826, versus $0.08184 for Opus 5.5. The calculation excludes cache charges, platform credits, tax, and markup.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Does Sonnet 5.5 beat Opus 5.5 on every benchmark?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"No. Anthropic\\u2019s own table is mixed by benchmark and effort setting: Sonnet 5.5 is higher on Terminal-Bench 4.0, while Opus 5.5 is higher on FrontierCode, CursorBench, GDPval-AA, Humanity\\u2019s Last Exam, OSWorld, and Chartography. These are provider-reported results, not an independent rerun.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Is Sonnet 5.5 a drop-in API replacement?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"No. The migration guide says thinking runs by default, thinking: disabled returns a 400 error, forced tool choice any\\\/tool is rejected, and clients should parse content blocks by type. Re-baseline token budgets and tool loops before switching.\"\n            }\n        },\n        {\n            \"@type\": \"Question\",\n            \"name\": \"Was this a long-context benchmark?\",\n            \"acceptedAnswer\": {\n                \"@type\": \"Answer\",\n                \"text\": \"No. The extraction prompt contains a short passage. It checks factual grounding and exact formatting; it does not measure behavior near the documented 1M-token context window.\"\n            }\n        }\n    ]\n}<\/script>","protected":false},"excerpt":{"rendered":"<p>Claude 5.5 comparison \u00b7 matched API tasks \u00b7 October 1, 2026 Claude Sonnet 5.5 vs Opus 5.5: Price, Benchmarks, and Real Tests Sonnet 5.5 and Opus 5.5 share the Claude 5.5 generation, but they are priced and positioned differently. This comparison puts both through the same five prompts and reports what the route actually returned. [&hellip;]<\/p>","protected":false},"author":13,"featured_media":20295,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_seopress_robots_primary_cat":"","_seopress_titles_title":"Claude Sonnet 5.5 vs Opus 5.5: Price & Real Tests","_seopress_titles_desc":"Claude Sonnet 5.5 vs Opus 5.5: compare official prices, benchmark caveats, local response time, token use, and five same-prompt API tasks.","_seopress_robots_index":"","footnotes":""},"categories":[7],"tags":[],"class_list":["post-20285","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-chat"],"acf":[],"_links":{"self":[{"href":"https:\/\/wp.glbgpt.com\/pt-br\/wp-json\/wp\/v2\/posts\/20285","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wp.glbgpt.com\/pt-br\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wp.glbgpt.com\/pt-br\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wp.glbgpt.com\/pt-br\/wp-json\/wp\/v2\/users\/13"}],"replies":[{"embeddable":true,"href":"https:\/\/wp.glbgpt.com\/pt-br\/wp-json\/wp\/v2\/comments?post=20285"}],"version-history":[{"count":3,"href":"https:\/\/wp.glbgpt.com\/pt-br\/wp-json\/wp\/v2\/posts\/20285\/revisions"}],"predecessor-version":[{"id":20337,"href":"https:\/\/wp.glbgpt.com\/pt-br\/wp-json\/wp\/v2\/posts\/20285\/revisions\/20337"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wp.glbgpt.com\/pt-br\/wp-json\/wp\/v2\/media\/20295"}],"wp:attachment":[{"href":"https:\/\/wp.glbgpt.com\/pt-br\/wp-json\/wp\/v2\/media?parent=20285"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wp.glbgpt.com\/pt-br\/wp-json\/wp\/v2\/categories?post=20285"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wp.glbgpt.com\/pt-br\/wp-json\/wp\/v2\/tags?post=20285"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}