Estimate from the version's published specs, benchmarks and reviews, 0-5; scores are only comparable within one criteria year. Scored as launched (Claude 3.7 Sonnet, 24 Feb 2025): 200K-token context (OpenRouter model page, released 24 Feb 2025; Willison notes 64K output, 128K in beta); price $3 in / $15 out per 1M tokens including thinking tokens = $6 blended at 3:1 (Anthropic); text and image input. Reasoning null: no launch-version Artificial Analysis Intelligence Index entry was opened (AA now re-scores on v4.1); Anthropic's headline results (SWE-bench Verified 63.7%, 70.3% with its scaffold) have no anchor. Coding null (no Terminal-Bench 4.0). Speed null (no tokens/s). Reviews: Zvi, Tom's Guide (7-prompt face-off vs GPT-4o, page last updated 2 May 2025, original date not shown), Willison, The Decoder.[1]
The base model is an iterative improvement and now you have access to optional reasoning capabilities. Claude 3.7 is especially good for coding.
Claude 3.7 Sonnet emerged as the clear winner, outperforming ChatGPT-4o.
As you may expect, 3.7 Sonnet is an improvement over 3.5 Sonnet - and is priced the same, at $3/million tokens for input and $15/m output.
Claude 3.7 Sonnet builds on an upgraded version of Claude 3.5 Sonnet but adds a significant new feature: an "Extended Thinking" mode that allows the model to reason through its responses.
| Claude 3.7 Sonnet and Claude Code · 2025年2月24日 | 0.50 |
|---|---|
| Anthropic launches Claude 3.7 Sonnet hybrid AI model and Claude Code programming tool · 2025年2月24日 | 0.50 |
| Time to Welcome Claude 3.7 · 2025年2月26日 | 0.70 |
| I put ChatGPT-4o vs Claude 3.7 Sonnet through a 7-round face-off — one left the other in the dust · 2025年4月29日 | 0.50 |
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。
该文章标注日期为“2025 年 2 月 24 日”:“今天,我们宣布推出 Claude 3.7 Sonnet”,并且它“现已在所有 Claude 套餐中可用”;TechCrunch[4] 称该模型“正于周一向所有用户和开发者推出”(2 月 24 日)。
Anthropic's[1] system card for "a hybrid reasoning model": it explains why users see the model's thinking, and states "Claude 3.7 Sonnet is released under the ASL-2 standard".
The Sonnet table dates Claude 3.7 Sonnet to 24 February 2025; the article adds that in February 2025 Claude 3.7 Sonnet playing Pokémon Red began streaming on Twitch to thousands of viewers.[1]
TechCrunch, 2025-02-24: the model "is rolling out to all users and developers on Monday"; free users get the standard mode, only paid plans the reasoning; $3/$15 per million tokens against o3-mini's $1.10/$4.40 and DeepSeek R1's $0.55/$2.19; "(Yes, the company skipped a number.)"
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 3.7 Sonnet Released
| 基准测试 | Claude 3.7 Sonnet(64K 扩展思考) | Claude 3.7 Sonnet(无扩展思考) | Claude 3.5 Sonnet(新版) | OpenAI o1¹ | OpenAI o3-mini¹(高强度) | DeepSeek R1(32K 扩展思考) | Grok 3 Beta(扩展思考) |
|---|---|---|---|---|---|---|---|
| 研究生水平推理(GPQA Diamond)³ | 78.2% / 84.8% | 68.0% | 65.0% | 75.7% / 78.0% | 79.7% | 71.5% | 80.2% / 84.6% |
| 智能体编程(SWE-bench Verified)² | — | 62.3% / 70.3% | 49.0% | 48.9% | 49.3% | 49.2% | — |
| 智能体工具使用(TAU-bench),零售 | — | 81.2% | 71.5% | 73.5% | — | — | — |
| 智能体工具使用(TAU-bench),航空 | — | 58.4% | 48.8% | 54.2% | — | — | — |
| 多语言问答(MMMLU) | 86.1% | 83.2% | 82.1% | 87.7% | 79.5% | — | — |
| 视觉推理(MMMU validation) | 75% | 71.8% | 70.4% | 78.2% | — | — | 76.0% / 78.0% |
| 指令遵循(IFEval) | 93.2% | 90.8% | 90.2% | — | — | 83.3% | — |
| 数学问题求解(MATH 500) | 96.2% | 82.2% | 78.0% | 96.4% | 97.9% | 97.3% | — |
| 高中数学竞赛(AIME 2024)³ | 61.3% / 80.0% | 23.3% | 16.0% | 79.2% / 83.3% | 87.3% | 79.8% | 83.9% / 93.3% |
在若干次试验(AIME 和 SWE-bench Verified 最多 16 次)上平均得出的 pass@1;第二个数字得益于并行测试时计算。¹ o1 的 TAU-bench 成绩由 OpenAI 公布后又被删除;TAU-bench 成绩可能不具可比性。² 62.3% 是在配合 bash/编辑器工具及一项“思考工具”的情况下,针对 500 个问题得出的 pass@1;70.3% 使用内部评分和在缩减子集上的定制脚手架;DeepSeek R1 使用 Agentless 框架。³ Claude 3.7 Sonnet 较高的 GPQA 和 AIME 2024 成绩使用了内部评分与并行测试时计算;o1 和 Grok 3 的成绩则使用了 64 个样本的多数投票法。
Epoch AI model database (Claude 3.7 Sonnet); Epoch's confidence in the figures: Likely. Blank tiles are not published.