Skip to content

Anthropic

Claude 3.7 Sonnet

Answers at once or thinks for longer first; Anthropic billed it as the first hybrid reasoning model. Launched with Claude Code.

Announced
Access
Public

Benchmark scores

TestScoreBest on the trackerHow it was measured
Humanity's Last Examexpert questions, no tools8%GPT-6 Astra 55%Independentleaderboard entry "Claude 3.7 Sonnet (Thinking)" (claude-3-7-sonnet-20250219_unknown)
OSWorld-Verifiedusing a computerPeople (original task set): 72%36%Claude Fable 5 86%Independentplain model, max 50 steps ("claude-3-7-sonnet-20250219", 129.34/361 tasks)Run 28 Jul 2025
GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated80%Gemini 4 Argon 96%Independent64K thinking budget (claude-3-7-sonnet-20250219_64K)Run 26 May 2025
ARC-AGI-2visual puzzles, model aloneAverage person: 60%Saturated0.9%GPT-6 Astra 95%IndependentARC Prize config "Claude 3.7 (8K)" (Claude 3.7 Thinking 8K)
OTIS Mock AIMEcompetition mathsSaturated58%GPT-5.5 100%Independent64K thinking budget (claude-3-7-sonnet-20250219_64K)Run 13 Mar 2025
SWE-bench Verifiedfixing real software bugsKnown flaws61%Claude Opus 4.7 83%Independentdefault settings (claude-3-7-sonnet-20250219)Run 4 Feb 2026
METR time horizontask length an AI agent finishes half the timeNo longer updated1 hClaude Mythos Preview 17 hIndependentMETR Horizon v1.1 run 'claude_3_7_sonnet_inspect'; 95% CI 33.0–104.2 min

Compare every model on the same tests →

Also on these charts

Sources