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
Partial outage now. Partially Degraded Service: Some users unable to sign up with Apple on the Claude iOS app
From the vendor's own status page. Uptime is the share of the window not covered by a major or critical incident (2727 min down, 658 min degraded); it covers the whole service, not this version alone.
Estimate from the version's published specs, benchmarks and reviews, 0-5; scores are only comparable within one criteria year. Scored as launched (Claude Sonnet 5.5, 28 Sep 2026): Artificial Analysis Intelligence Index 56 on v4.3 (max effort, default fallback). Terminal-Bench 4.0 64% as measured by Artificial Analysis (Anthropic's own run: 70.6%). AA tested a pre-release build with a structured-output bug and expects minimal or slightly understated change. 1M-token context; $2 in / $10 out per 1M (about $4 blended). 132.7 output tokens/s on Artificial Analysis (live page, 24 Oct 2026; Anthropic claims 30%+ faster than Sonnet 5). Text and image input. 5 reviews, string-matched.[1]
Claude Sonnet 5.5 (max) makes large strides on Terminal-Bench, sitting among the top models for both Terminal-Bench 4.0 and Terminal-Bench-Science.
Sonnet 5.5 suffered from the same bug as Opus 5.5 : the "max" thinking effort pelican thought for 128,000 tokens (at a cost of $1.28) before running out of tokens and failing to produce an SVG.
On this particular test, at least, the cheaper Sonnet has gone from trailing Opus by a wide margin to essentially tying it.
Sonnet 5.5's gaps against Opus 5.5 show up more clearly outside agentic and knowledge-work tasks.
At the highest effort level, "Max," Sonnet 5.5 actually scores worse on FrontierCode than at "Xhigh."
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。
Anthropic 的公告标题为“Claude[4] Sonnet 5.5 2026 年 9 月 28 日”,文中写道该模型“现已在所有平台上可用”;模型页面列出的发布日期为“2026 年 9 月 28 日”,TechCrunch[2](2026 年 9 月 28 日太平洋夏令时上午 11:00)和 CNBC[3](“Anthropic 周一发布”)的说法与此一致。
TechCrunch's same-day report by Lucas Ropek (11:00 AM PDT, September 28, 2026): Anthropic[1] "has released the newest version of Sonnet", about three months after Sonnet 5, with the same cyber safeguards as Fable and Opus, and a new Haiku "in the coming weeks" with no firm date.
CNBC: "Anthropic[1] on Monday released a new AI model called Sonnet 5.5", its second launch since Dario Amodei urged labs to slow frontier progress, at $2 and $10 per million tokens, "half the price of Opus 5.5", with Haiku 5.5 "coming soon".
Anthropic's[1] model page: claude-sonnet-5-5, 1M-token context, 128K max output (300K on the Batch API in beta), adaptive thinking with high default effort, text and images in, a June 2026 knowledge cutoff, cache writes $2.50 and reads $0.20 per million tokens, and five breaking changes from Sonnet 5.
SiliconANGLE's same-day report adds that Sonnet 5.5 carries Anthropic's[1] invisible text watermarking, added to comply with rules including the EU AI Act, and that Haiku 5.5 is planned "in the coming weeks".
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 Sonnet 5.5 Released
claude-sonnet-5-5 的名称使用,并登陆 Amazon Web Services、Google Cloud 和 Microsoft Azure(零数据留存),以及 GitHub Copilot[1][4][6]| 基准测试 | Sonnet 5.5 | Sonnet 5 | Opus 5.5 | GPT-6 Sol |
|---|---|---|---|---|
| Terminal-Bench 4.0(智能体编程) | 70.6% | 10.3% | 66.4%¹ | — |
| FrontierCode 1.1(主项)(智能体编程) | 46.2%(Max²)、52.1%(Xhigh) | 42.4% | 54.4% | 49.3% |
| CursorBench 4.0(智能体编程) | 55.5% | 34.1% | 57.8% | — |
| GDPval-AA v2.1³(知识型工作) | 1844 | 1449 | 1846 | 1487⁴ |
| AA-Briefcase v1.1³(知识型工作) | 1811 | 1359 | 1822 | 1483⁴ |
| Humanity's Last Exam(跨学科推理) | 64.5% 使用工具 | 54.9% 使用工具 | 67.7% 使用工具 | — |
| OSWorld 2.1(计算机操作) | 80.1% 部分 | 57.0% 部分 | 81.8% 部分 | — |
| Chartography(图表视觉识别) | 61.6% 不使用工具 | 15.6% 不使用工具 | 64.4% 不使用工具 | 53.6%⁴ 不使用工具 |
¹ Opus 5.5 在 Xhigh 强度下的成绩,是其最高分。² Sonnet 5.5 在 Max 强度下的 FrontierCode 得分低于 Xhigh 强度,因为该基准会对超出范围的改动扣分。³ 由 Artificial Analysis 在 Sonnet 5.5 一个存在已修复的结构化输出错误的预发布版本上运行。⁴ GPT-6 Sol 的成绩可能早于 OpenAI 修复其图像理解能力的错误之前。