0

Claude 4 Sonnet

Claude 4 Sonnet is an AI model from Anthropic.

Publisher
Anthropic
Family
Claude
Context
200K
Input
Output
License
proprietary
Released
22 May 2025

Cite

Notes

Only stored in your browser.

Attribution

Attribution policy →
Context
200K
Weights
Closed
API only
Input
$0
/1M tok
Output
$0
/1M tok
Blended
$0
#1/ 499/1M · 3:1
Speed
0
#103/ 498tok/s

Price & speed via Artificial Analysis · ranked across tracked models

Reasoning effortIndexSpeed
Reasoning18.90 t/s
Non-reasoning16.60 t/s

Intelligence Index per reasoning-effort setting, via Artificial Analysis · pricing identical across tiers

Reported on 40 evals across 12 domains - best 99.1% on MATH-500 · #4 of 190

Reported eval scores

40
Wiki Race84
Twenty Questions0.6
Mini Swe Agent Bench86.7
Capa47.8
Battleship22.9
SWE-bench Lite58.3
SWE-bench Verified74.6
LiveCodeBench65.5
MATH-50099.1
AIME 2024: Problems from the American Invitational Mathematics Examination77.3
MMLU-Pro84.2
LiveBench64.8
Humanity's Last Exam (HLE)10.7
LiveBench - Coding77.5
SWE-bench Multimodal35.6
LiveBench - Language72.9
Terminal-Bench (Hard)31.1
IFBench54.7
LiveBench - Reasoning69
LiveBench - Data Analysis54.6

Head-to-heads

29

benchmark comparisons against other models

vs GPT-5 Mini29 shared evalsvs Claude Sonnet 4.528 shared evalsvs Claude Opus 4.526 shared evalsvs GPT-525 shared evalsvs GPT-5 Nano25 shared evalsvs GPT-5.225 shared evalsvs Gemini 2.5 Pro25 shared evalsvs Grok 4.1 Fast24 shared evalsvs GPT-5.523 shared evalsvs Gemini 3.1 Pro Preview23 shared evalsvs Kimi K2.523 shared evalsvs Claude Opus 4.822 shared evalsvs Claude Fable 522 shared evalsvs Claude Opus 4.622 shared evalsvs GPT-5.422 shared evalsvs GLM 4.722 shared evalsvs Claude Opus 4.721 shared evalsvs DeepSeek V3.221 shared evalsvs gpt-oss-120b21 shared evalsvs GLM 5.120 shared evalsvs Claude Sonnet 4.620 shared evalsvs GPT-4o20 shared evalsvs GPT-4.1 Mini19 shared evalsvs GPT-4.119 shared evalsvs Qwen3.6 Plus19 shared evalsvs GPT-5.4 Mini19 shared evalsvs GPT-4o-mini18 shared evalsvs GPT-4.1 Nano16 shared evalsvs GPT-6 Astra8 shared evals