See key specs and per-benchmark scores for each model/mode. Scroll horizontally for all columns. 当前对比 2 个模型的评测数据与核心参数。
Claude Sonnet 4.5
Anthropic
Each axis is a category average, normalized to a 100-point radar.
Relative edge: Agent能力评测 +29.4 / Relative gap: none clear
Relative edge: none clear / Relative gap: Agent能力评测 -29.4
Method: for each model and benchmark, the chart first averages all scores in the current mode scope instead of taking the best score, then averages those benchmark scores within each category. Only benchmarks with at least two selected models scored are included; missing values are not counted as zero.
Best overall
Claude Sonnet 4.5 · 60.59
Best single
Claude Sonnet 4.5 · AIME2025 100.00
Modality coverage
Claude Sonnet 4.5 · 2 modalities
Head to head
24
Benchmarks
21
Wins
1
Losses
+9.21
Average diff
Compare benchmark results across thinking modes and tool usage.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
Complete scores for each model/mode across selected benchmarks.
24 benchmarks with comparable scores. Each model shows its best score; mode label is displayed below.
| Benchmark | Claude Sonnet 4.5 | Claude Sonnet 4 |
|---|---|---|
ARC-AGI 综合评估 | 63.70Thinking Enabled | 40.00Thinking Enabled |
ARC-AGI-2 综合评估 | 13.60Thinking Enabled | 5.90Thinking Enabled |
GPQA Diamond 综合评估 | 83.40Thinking Enabled | 83.80Deep Thinking Mode ｜ Tools |
HLE 综合评估 | 33.60Thinking Enabled ｜ Tools | 9.60Thinking Enabled |
LiveBench 综合评估 | 78.26Thinking Enabled | 73.82Thinking Enabled |
MMLU Pro 综合评估 | 88.00Thinking Enabled | 84.00Thinking Enabled |
LiveCodeBench 编程与软件工程 | 71.00Thinking Enabled | 66.00Thinking Enabled |
SWE-Bench Pro - Public 编程与软件工程 | 43.60Thinking Enabled | 42.70Thinking Enabled |
SWE-bench Verified 编程与软件工程 | 77.20Thinking Enabled ｜ Tools | 72.70Thinking Enabled ｜ Tools |
AIME2025 数学推理 | 100.00Thinking Enabled ｜ Tools | 85.00Deep Thinking Mode ｜ Tools |
FrontierMath 数学推理 | 5.20Standard Mode | 4.10Standard Mode |
IMO-ProofBench 数学推理 | 27.10Thinking Enabled | 27.10Thinking Enabled |
4.80Thinking Enabled | 4.80Thinking Enabled | |
OSWorld-Verified AI Agent - 工具使用 | 61.40Thinking Enabled ｜ Tools | 42.20Thinking Enabled ｜ Tools |
Terminal-Bench AI Agent - 工具使用 | 50.00Thinking Enabled ｜ Tools | 41.30Deep Thinking Mode ｜ Tools |
MMMU 多模态理解 | 77.80Thinking Enabled | 76.50Standard Mode |
Simple Bench 常识推理 | 54.30Standard Mode | 45.50Thinking Enabled |
τ²-Bench Agent能力评测 | 84.70Thinking Enabled ｜ Tools | 52.00Standard Mode ｜ Tools |
τ²-Bench - Telecom Agent能力评测 | 98.00Thinking Enabled ｜ Tools | 65.00Thinking Enabled ｜ Tools |
IF Bench 指令跟随 | 57.30Thinking Enabled ｜ Tools | 55.00Thinking Enabled ｜ Tools |
GDPval-AA 生产力知识 | 39.00Thinking Enabled | 33.00Thinking Enabled |
AA-LCR 长上下文能力 | 66.00Thinking Enabled | 65.00Thinking Enabled |
Claw Bench OpenClaw智能体能力综合测评 | 88.10Thinking Enabled ｜ Tools | 77.80Thinking Enabled ｜ Tools |
Pinch Bench OpenClaw智能体能力综合测评 | 88.20Thinking Enabled ｜ Tools | 80.50Thinking Enabled ｜ Tools |
Side-by-side input/output token pricing
Licensing, MoE architecture, and multi-modality support.
| Features & specs | Claude Sonnet 4.5Anthropic | Claude Sonnet 4Anthropic |
|---|---|---|
Core specsRelease | 2025-09-30 | 2025-05-23 |
Context length | 1000K | 200K |
Max output | 65536 | 64000 |
MoE | No | No |
Supported modes | 常规模式（Non-Thinking Mode）思考模式（Thinking Mode）深度思考（Deeper Thinking Mode） | 常规模式（Non-Thinking Mode）思考模式（Thinking Mode）深度思考（Deeper Thinking Mode） |
LicenseCode Open Source | Not provided | Not provided |
Weights Open Source | Not provided | Not provided |
Commercial use | 不开源 | 不开源 |
Modality supportText Input/Output | / | / |
Image Input/Output | / | / |
ResourcesPaper / report | Introducing Claude Sonnet 4.5 | Introducing Claude 4 |
DataLearner blog | 全球最强编程大模型升级：Anthropic发布Claude Sonnet 4.5！同时还有一波重磅工具更新：Claude Code支持保存状态等 | Anthropic发布Claude4，全球最强编程大模型，大幅提升AI Agent系统所需的各项能力，最长可以7小时连续工作，80%工程任务自主完成 |
Claude Sonnet 4
Anthropic
