Anthropic
Anthropic

Claude Opus 4

Release Date 2025-05-22(Knowledge cutoff 2025-05-01)1M IN · 128K OUTProprietary Model

Claude Opus 4 was Anthropic's model for difficult coding and extended tool workflows, and is now retired on Anthropic-operated APIs. Anthropic emphasized more sustained work than earlier Claude models, combining responsive answers or extended reasoning with external tools as needed. When given file access, it could record important facts for later work, but this depended on storage tools and did not guarantee unchanged performance or complete context retention throughout a long task. Current AIB comparisons place it on the weaker side for documents and automation, distinct from its position in launch benchmarks. New work should follow current Opus successor and migration guidance; partner clouds may have different retirement schedules, so actual availability, code behavior and analytical evidence still need checking.

API cost

Official price
$15IN$75OUT
SourceAnthropic official docs

Output Speed

Very slowper 2,000 tokens
34tok/s
First Output 1.3stotal: 60.2s
SourceArtificial AnalysisLast updated

User Comparisons

Response Speed
Total 13.3s
45 tok/s
SourceAIB Inc.Last updated

Expert Performance Evaluation

142687.3%20.6

More about Arena Text

73.4%1h 40m

More about TAU2

69.3%140961.1%43.6%

More about Long Context Reasoning

46.8%Reasoning effort: Budget 2K112712.0%88.0%60.6

More about Deep Research Bench

149831.1%34.063.6%39.8%70.7%43.7%6.9%72.0%14681.33×

More about Arena Coding

1432143953.7%143314448.361422

More about Arena Writing, Literature & Language

12.3%0.3%142179.6%64.4%73.373.3%1438142698.2%85.0%4.5%4.2%35.7%8.6%58.8%5.6%81.6%37.4

By effortNone 6.2%Thinking 12.3%

More about Humanity's Last Exam

119238.0%49.0%

More about Arena Vision

Specialized fields

13851387145714501441143338.0%88.0%73.6%79.6%67.6%87.0%84.2%94.2%78.9%86.0%84.0%66.7%35.0%13.0%79.0%87.9%96.0%72.0%73.0%95.0%71.0%78.0%88.8%73.0%6.0%25.1%

More about Arena Korean

OddBench

Does it call out a nonsense question instead of playing along?BullshitBench v2Bottom 41%#86
34.0%

More about BullshitBench v2

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