Claude Sonnet 4.5vsClaude Sonnet 4
在 21 个共同 benchmark 中，Claude Sonnet 4.5 整体领先：Claude Sonnet 4.5 领先 18 项，Claude Sonnet 4 领先 1 项，持平 2 项，平均分差 +22.01。
Claude Sonnet 4.5
Anthropic · 2025-09-30 · 聊天大模型
Claude Sonnet 4
Anthropic · 2025-05-23 · 推理大模型
Claude Sonnet 4.518 项(86%)持平2(5%)1 项Claude Sonnet 4
评测分数
按能力类目分组，每组内按分差大小排列；共 21 项。
Coding and Software Engineer
Claude Sonnet 4.5 领先 4/4| 评测项 | Claude Sonnet 4.5 | Claude Sonnet 4 | 分差 |
|---|---|---|---|
| CodeClash | 1,3891 / 8Normal (With Tools) | 1,2234 / 8Normal (With Tools) | +166 |
| LiveCodeBench | 5977 / 128Normal (No Tools) | 48.50170 / 250Normal (No Tools) | +10.50 |
| SWE-bench Verified | 828 / 116Parallel Thinking (With Tools) | 80.2014 / 116Parallel Thinking (With Tools) | +1.80 |
| SWE-Bench Pro - Public | 43.6054 / 62Thinking (No Tools) | 42.7055 / 62Thinking (No Tools) | +0.90 |
General Knowledge
Claude Sonnet 4.5 领先 4/4| 评测项 | Claude Sonnet 4.5 | Claude Sonnet 4 | 分差 |
|---|---|---|---|
| MMLU Pro | 887 / 134Thinking (No Tools) | 8439 / 175Thinking (No Tools) | +4 |
| LiveBench | 53.6985 / 117Normal (No Tools) | 50.9891 / 117Normal (No Tools) | +2.71 |
| ARC-AGI-1 | 25.50127 / 147Normal (No Tools) | 23.80128 / 147Normal (No Tools) | +1.70 |
| HLE | 7.10184 / 197Normal (No Tools) | 5.52467 / 563Normal (No Tools) | +1.58 |
Math and Reasoning
胶着 4/4| 评测项 | Claude Sonnet 4.5 | Claude Sonnet 4 | 分差 |
|---|---|---|---|
| FrontierMath | 5.2038 / 60Normal (No Tools) | 4.1041 / 60Normal (No Tools) | +1.10 |
| AIME2025 | 37107 / 117Normal (No Tools) | 38170 / 215Normal (No Tools) | -1 |
| IMO-ProofBench | 27.108 / 16Thinking (No Tools) | 27.108 / 16Thinking (No Tools) | 持平 |
| IMO-ProofBench Advanced | 4.8019 / 24Thinking (No Tools) | 4.8019 / 24Thinking (No Tools) | 持平 |
AI Agent - Tool Usage
Claude Sonnet 4.5 领先 2/2| 评测项 | Claude Sonnet 4.5 | Claude Sonnet 4 | 分差 |
|---|---|---|---|
| OSWorld-Verified | 61.4022 / 26Thinking (With Tools) | 42.2024 / 26Thinking (With Tools) | +19.20 |
| Terminal-Bench | 2725 / 35Normal (With Tools) | 2626 / 35Normal (With Tools) | +1 |
Claw-style Agent Evaluation
Claude Sonnet 4.5 领先 2/2| 评测项 | Claude Sonnet 4.5 | Claude Sonnet 4 | 分差 |
|---|---|---|---|
| Claw Bench | 88.1013 / 29Thinking (With Tools) | 77.8023 / 29Thinking (With Tools) | +10.30 |
| Pinch Bench | 88.205 / 38Thinking (With Tools) | 80.5023 / 38Thinking (With Tools) | +7.70 |
Agent Level Benchmark
Claude Sonnet 4.5 领先 1/1| 评测项 | Claude Sonnet 4.5 | Claude Sonnet 4 | 分差 |
|---|---|---|---|
| τ²-Bench | 7126 / 44Normal (With Tools) | 5235 / 44Normal (With Tools) | +19 |
Long Context
Claude Sonnet 4.5 领先 1/1| 评测项 | Claude Sonnet 4.5 | Claude Sonnet 4 | 分差 |
|---|---|---|---|
| AA-LCR | 54133 / 170Normal (No Tools) | 44150 / 170Normal (No Tools) | +10 |
Productivity Knowledge
Claude Sonnet 4.5 领先 1/1| 评测项 | Claude Sonnet 4.5 | Claude Sonnet 4 | 分差 |
|---|---|---|---|
| GDPval-AA | 3910 / 15Thinking (No Tools) | 3313 / 15Thinking (No Tools) | +6 |
Writing and Creative Capabilities
Claude Sonnet 4.5 领先 1/1| 评测项 | Claude Sonnet 4.5 | Claude Sonnet 4 | 分差 |
|---|---|---|---|
| Creative Writing | 1,67430 / 106Normal (No Tools) | 1,48053 / 106Normal (No Tools) | +194.10 |
科学与综合推理
Claude Sonnet 4.5 领先 1/1| 评测项 | Claude Sonnet 4.5 | Claude Sonnet 4 | 分差 |
|---|---|---|---|
| GPQA Diamond | 73.70179 / 274Normal (No Tools) | 68337 / 462Normal (No Tools) | +5.70 |
规格对比
| 字段 | Claude Sonnet 4.5 | Claude Sonnet 4 |
|---|---|---|
| 发布机构 | Anthropic | Anthropic |
| 发布时间 | 2025-09-30 | 2025-05-23 |
| 模型类型 | 聊天大模型 | 推理大模型 |
| 架构 | 稠密模型 | 稠密模型 |
| 参数规模 | 暂无数据 | 暂无数据 |
| 上下文长度 | 1000K | 200K |
| 最大输出 | 64K | 64K |
API 调用价格
价格优先使用 DataLearner 配置的 API 记录；缺失项不做推测。
| 价格项 | Claude Sonnet 4.5 | Claude Sonnet 4 |
|---|---|---|
| 文本输入 | $3 / 1M tokens | $3 / 1M tokens |
| 文本输出 | $15 / 1M tokens | $15 / 1M tokens |
| 缓存读取 | $0.3 / 1M tokens | $0.3 / 1M tokens |
| 缓存写入 | $3.75 / 1M tokens | $3.75 / 1M tokens |
小结
- Claude Sonnet 4.5在以下类目领先:Coding and Software Engineer (4/4)、General Knowledge (4/4)、AI Agent - Tool Usage (2/2)、Claw-style Agent Evaluation (2/2)、Agent Level Benchmark (1/1)、Long Context (1/1)、Productivity Knowledge (1/1)、Writing and Creative Capabilities (1/1)、科学与综合推理 (1/1)
- 胶着类目:Math and Reasoning
21 个共同 benchmark 上，Claude Sonnet 4.5 平均高出 22.01 分。
单项差距最大的 benchmark：Creative Writing — Claude Sonnet 4.5 1,674，Claude Sonnet 4 1,480（分差 +194.10）。
本页正文由结构化模型、价格与 benchmark 数据生成，不使用实时 LLM 撰写。