Claude Sonnet 5
Anthropic·5 个数据来源·更新于 2026-11-10
评测表现
只在同一来源、同一口径下比较。切换来源可查看不同测试结果。
不同来源的分数不能直接相加或横向比较;评测任务成本不是 API Token 单价。
逐项基准成绩
24 项已收录成绩 · 保留原始单位和测试条件
Research-level final-answer math (MathArena, April + May 2026 releases, 81 problems). With tools: 72.2%. Without tools: 65.7%. Extended thinking, averaged over 4 attempts per problem.
成绩出处Zapier end-to-end business-workflow benchmark, private held-out leaderboard set. Max effort.
成绩出处BenchCAD Vision2Code voxel IoU. With Python tools: 0.373. Without tools: 0.266. Random 1,000-file subset averaged over 5 runs.
成绩出处Agentic search. Single-agent with web search, web fetch, code execution, adaptive thinking at max effort, 10M-token limit, context compaction at 200k tokens. Multi-agent reaches 86.6%.
成绩出处Chart QA requiring visual reasoning (1,162 questions). With Python tools: 86.7%. Without tools: 70.1%. Test split averaged over 5 runs, Claude Sonnet 4.6 grader.
成绩出处CharXiv Reasoning (1,000 validation questions). With Python tools: 88.3%. Without tools: 77.0%. Averaged over 5 runs, Claude Sonnet 4.6 grader.
成绩出处DeepSWE v1.1 leaderboard, mini-swe-agent harness, max effort; Pass@1 54% ± 4%.
成绩出处FrontierCode v1, Cognition's production-quality coding evaluation, at max effort.
成绩出处FrontierCode 1.1 current leaderboard; mergeability score at xhigh effort: 42.7%.
成绩出处Expert multimodal document QA (Surge AI, 100 prompts/PDFs). Mean criteria pass rate with Python tools: 81.6%. Without tools: 67.5%. Internal harness, averaged over 5 runs.
成绩出处Length-adjusted score with adaptive thinking at max effort. Claude Opus 4.8 grader, averaged over 5 trials.
成绩出处Multidisciplinary reasoning (2,500 questions). With tools (web search, web fetch, code execution): 57.4%. Without tools: 43.2%. Auto thinking, 1M total-token cap, Claude Opus 4.6 grader.
成绩出处All-pass rate on Harvey's held-out set (mean criterion-pass rate 91.2%). Full public set all-pass rate: 8.9% (88.26% mean criterion-pass). Adaptive thinking at max effort.
成绩出处Strict pass rate on 55 regulatory-affairs tasks. Plain API via Vercel AI Gateway with a web_search tool (Exa), provider default settings, one answer per task, graded by GPT-6.1 Sol at medium reasoning effort (Arca, 4 Nov 2026 edition).
成绩出处Grounded reasoning over U.S. Treasury Bulletin documents (133-question subset). Internal agentic harness with extracted-text documents and code-execution tools. Exact-match, mean of 5 trials. OfficeQA full set: 73.3%.
成绩出处Agentic computer use. 361 tasks at 100 steps, 1080p resolution, adaptive thinking at max effort. First-attempt success rate averaged over 5 runs. Updated harness (zoom-tool bug fix, 128K max tokens per turn).
成绩出处300 problems across 9 programming languages. Averaged over 5 trials at max effort.
成绩出处Visual context (screenshots, design mockups) added to issue descriptions. Internal harness, averaged over 5 trials.
成绩出处Harder SWE-bench variant: actively-maintained repos with larger multi-file diffs and reduced public ground-truth leakage. Averaged over 5 trials at max effort.
成绩出处Agentic coding (SWE-bench Verified, 500-problem subset). Standard configuration, adaptive thinking at max effort, thinking blocks included in sampling, averaged over 5 trials.
成绩出处Agentic terminal coding (Terminal-Bench 2.1) with the mini-SWE-agent harness at xhigh effort. Mean reward over 5 attempts across 89 unique tasks (445 trials).
成绩出处Terminal-Bench 4.0 public leaderboard; resolution rate 12.4% ± 3.1%; Claude Code agent; max effort.
成绩出处Pass@1 averaged over 3 trials across all 108 tasks. Internal harness with adaptive thinking at max effort. Pass@3: 63.0%, Pass^3: 40.7%.
成绩出处Proof-based math (USAMO 2026, held March 21-22 2026). MathArena grading methodology (proofs rewritten by a neutral model and judged by a 3-model panel; minimum score taken). High effort, 300k token limit, averaged over 10 attempts. 79.5%.
成绩出处技术规格
只展示有可靠来源的资料;优先采用厂商官网与官方模型仓库。
价格与用量
美元 / 百万 Token。不同服务商或测试配置的报价分别列出。
| 来源 / 服务商 | 输入 | 输出 | 缓存读取 |
|---|---|---|---|
| 厂商官网 · Claude Sonnet 5 | $2 | $10 | $0.2 |
| Anthropic | $2 | $10 | $0.2 |
| DeepInfra | $3 | $15 | 待补充 |
每个数字,都有出处
同名模型通过开发机构和明确版本关联。预览版、日期版本和不同规模模型分别建档。
厂商官网
厂商官方文档 · 基础规格与标准 API 报价
LLM Stats
公开模型页与公开榜单;自报成绩单独标识
BenchLM
公开评测数据;保留来源原始口径
Arena
Arena 官方公开数据集 · CC BY 4.0
OpenRouter
公开模型目录与 API 报价
查看全部配置与采集记录
| 来源 / 配置 | 采集时间 | 原始数据 |
|---|---|---|
| LLM StatsClaude Sonnet 5SHA-256 38986ec9b954a818… | 2026-11-10T13:53:09Z | 查看来源 |
| BenchLMClaude Sonnet 5 | 2026-11-10 | 查看来源 |
| ArenaClaude Sonnet 5 (High) | 2026-11-10T14:07:52Z | 查看来源 |
| ArenaClaude Sonnet 5 (High) | 2026-11-10T14:07:52Z | 查看来源 |
| ArenaClaude Sonnet 5 (High) | 2026-11-10T14:07:52Z | 查看来源 |
| ArenaClaude Sonnet 5 (High) | 2026-11-10T14:07:52Z | 查看来源 |
| ArenaClaude Sonnet 5 (High) | 2026-11-10T14:07:52Z | 查看来源 |
| OpenRouterClaude Sonnet 5SHA-256 63cbc45064d84060… | 2026-11-10T13:51:49Z | 查看来源 |
| 厂商官网Claude Sonnet 5SHA-256 188e8eb2ba29b33c… | 2026-11-10T14:06:47Z | 查看来源 |