真实、承认不确定性、说明假设、指出细微差别、不谄媚;为在所有被评模型中的百分位排名。Artificial Analysis发布文章,结果截至2026年10月19日(实时榜单已增至41个模型,数值可能不同)。无系统提示;900道非公开问题;由Gemini 3.8 Flash、Claude Sonnet 5.5和GLM 5.3 Flash两两评判。衡量的是特质的强度,而非是否可取,也不是能力基准。 · 来源
友好特质的两两比较Elo;锚定模型GPT-6.1 Sol固定为1000。Artificial Analysis发布文章,结果截至2026年10月19日(实时榜单已增至41个模型,数值可能不同)。无系统提示;900道非公开问题;由Gemini 3.8 Flash、Claude Sonnet 5.5和GLM 5.3 Flash两两评判。衡量的是特质的强度,而非是否可取,也不是能力基准。 · 来源
| 模型 | 透明度指数 | 友好Elo(95%置信区间) | 每次回答字数 | 每次回答列表数 | 每次回答加粗数 |
|---|---|---|---|---|---|
| Claude Sonnet 5.5 (最高(含回退)) | 100 | 1041 (1010-1070) | 450 | 2.84 | 9.95 |
| Claude Fable 5.1 (最高(含回退)) | 97 | 1047 (1014-1076) | 576 | 1.73 | 8.67 |
| GPT-6 Astra (最高) | 94 | 1001 (972-1031) | 352 | 1.46 | 8.84 |
| Claude Opus 5.5 (最高(含回退)) | 88 | 1077 (1049-1107) | 571 | 2.92 | 11.8 |
| Gemini 4 Argon (高) | 18 | 1059 (1030-1087) | 790 | 3.88 | 17.2 |
字数、列表数和加粗数为900个问题中每次回答的平均值,仅作描述,并非质量评分。Artificial Analysis发布文章,结果截至2026年10月19日(实时榜单已增至41个模型,数值可能不同)。无系统提示;900道非公开问题;由Gemini 3.8 Flash、Claude Sonnet 5.5和GLM 5.3 Flash两两评判。衡量的是特质的强度,而非是否可取,也不是能力基准。 · 来源
Estimate from the version's published specs, benchmarks and reviews, 0-5; scores are only comparable within one criteria year. Scored as launched (Claude Opus 5.5, 22 Sep 2026): Artificial Analysis Intelligence Index 58 on v4.3 (max effort, default fallback; #1). Terminal-Bench 4.0 59.6% as measured by Artificial Analysis (Anthropic's own run: 66.4%, different harness). 1M-token context; $4 in / $20 out per 1M (about $8 blended; the pin text notes the press cites the old $5/$25). 96.3 output tokens/s on Artificial Analysis (live page, 24 Oct 2026). Text and image input. 4 reviews, string-matched (one is a NYU summary of AA's result).[1]
Claude Opus 5.5 brings Anthropic to parity with GPT-6 Astra on evaluations like Terminal-Bench 4.0 and AutomationBench-AA, while extending Anthropic's lead in agentic knowledge work.
In a first for my " Generate an SVG of a pelican riding a bicycle " test, Claude Opus 5.5 at "max" thinking level failed to return a response!
Claude Opus 5.5 consistently impresses me when a task requires judgment or when I need my messy notes or thoughts to be turned into something useful.
Readers choosing a model for terminal-heavy agent work will see a smaller difference than the headline score suggests.
读者上传的照片,与图钉带来源的媒体分开显示。你可以为吃到或看到的内容评分,在餐厅图钉上还能直接在照片上标记菜品。
第一项始终是图钉的来源。总体可信度是各份资料对上方所用开始和结束时间支持程度的加权平均;资料每比最新的一份旧 180 天,权重减半。
与Fable 5.1的公告不同,该页面自带日期:标题栏写着“Claude[2] Opus 5.5 September 22, 2026”。TechCrunch[3]的署名时间(“9:30 AM PDT · September 22, 2026”、“于周二发布”)以及XDA的2026-09-22T16:44Z时间戳均一致,因此来源维持该说法,两篇报道也相互印证。
Anthropic's[1] Claude Opus 5.5 model page lists a 1M-token context window, 128K max output (300K Batch beta), adaptive thinking always on with medium default effort, and a Jun 2026 knowledge cutoff.
TechCrunch's same-day report by Russell Brandom, bylined “9:30 AM PDT · September 22, 2026” and saying the model “was released on Tuesday”. It supplies the Opus 5 comparison prices ($25 per million output tokens) that the announcement states only as a percentage.
XDA's report, timestamped 2026-09-22T16:44Z, independently confirms the launch day and gives both old prices in full - “Claude[2] Opus 5 costs $5 for 1 million input tokens and $25 for 1 million output tokens” against $4 and $20.
有遗漏或错误吗?用你自己的话说明:能佐证此图钉的链接、不同的开始或结束日期及理由,或缺失、有误的信息。AI 会对照此图钉的来源进行核实,搜索更好的来源,并添加任何支持你说法的页面。图钉自身的来源仍然最重要。AI 也会查看图片:显示的是别的东西或显示效果差的图片会被移到后面或替换。
原标题: Claude Opus 5.5 Released
claude-opus-5-5;Claude Sonnet 5.5和Claude Haiku 5.5"将在未来几周内推出"[1]