Anthropic
Anthropic

Claude Opus 4.1

Release Date 2025-08-05(Knowledge cutoff 2025-01-31)200K IN · 32K OUTProprietary ModelAPI available

Claude Opus 4.1 was Anthropic's update to an earlier Opus for coding, reasoning and tool use; it has now been retired on Anthropic-operated APIs. Anthropic highlighted gains in changes spanning multiple files and in research and analysis that require tracking details, including narrowing down the parts of a codebase that need repair. Those launch evaluations and examples do not guarantee that every edit is correct or free of unnecessary changes. Current AIB comparisons place it in the middle of the pack for documents and on the weaker side for research and automation. New integrations should use current successor and migration guidance, while existing users should check their actual provider because partner-cloud retirement schedules can differ, and still validate code and analytical results.

API cost

Official price
$15IN$75OUT
SourceAnthropic official docs

Output Speed

Very slowper 2,000 tokens
15tok/s
First Output 3.4stotal: 137s ›
SourceOpenRouterLast updated

Expert Performance Evaluation

145088.0%22.8

More about Arena Text

71.4%1h 54m

More about TAU2

76.0%144743.6%47.2%50.9%

More about Long Context Reasoning

143548.3%114811.8%88.2%60.1

More about Arena Text Factuality

151167534.3%36.565.4%40.9%73.3%45.9%149238.0%1.34×

More about Arena Coding

1446147155.4%144614588.4773.9%1435

More about Arena Writing, Literature & Language

12.5%144480.9%68.9%80.380.3%12.6%2.4%147014517.2%4.2%60.0%7.2%21.0%80.0%37.1

More about Humanity's Last Exam

Can it predict how objects will move just from a picture?VPCTBottom 30%#8
35.0%

More about VPCT

Specialized fields

14221412148014821476146342.0%7.9%54.8%33.1%33.2%65.0%33.1%94.3%94.0%95.0%90.0%75.9%80.0%71.8%93.0%85.0%94.1%82.1%83.0%83.0%70.0%46.0%16.0%83.0%88.9%92.0%74.0%72.0%99.0%66.0%29.9%79.9%89.8%75.0%31.0%8.0%25.3%

More about Arena Korean

OddBench

Does it call out a nonsense question instead of playing along?BullshitBench v2Bottom 50%#72
43.0%Reasoning effort: None

By effortNone 43.0%High 42.0%

More about BullshitBench v2

See each evaluation’s detail page for its source and last updateData corrections & removal requests