Anthropic
Anthropic

Claude Opus 4.5

Release Date 2025-11-24(Knowledge cutoff 2025-08-01)200K IN · 64K OUTProprietary ModelAPI available

Claude Opus 4.5 is Anthropic's model for complex software engineering, tool-assisted research and document work. Anthropic reported improvements over earlier Opus models in coding, computer use and token efficiency, with an effort setting to balance speed, cost and reasoning depth. It can be connected to browser and spreadsheet workflows, but file editing and record management depend on the surrounding product and execution environment. Current AIB comparisons place it in the middle of the pack for coding and documents and on the weaker side for automation, rather than preserving its position in launch-time evaluations. Complex requirements can still be missed and tool actions can be wrong, so users should verify results, run code tests and compare actual task time and cost with successor models.

API cost

Official price
$5IN$25OUT
SourceAnthropic official docs

Output Speed

Very slowper 2,000 tokens
43tok/s
First Output 1.1stotal: 47.6s
SourceOpenRouterLast updated

Expert Performance Evaluation

147089.5%72.629.1

More about Arena Text

$4,96789.5%62.3%3.8%4h 53m66.3%84.0%79.6%

More about Vending-Bench 2

77.3%147245.5%65.1%49.2%146674.458.8%55.2%21.1%

By effortNone 70.7%Thinking 77.3%

More about Long Context Reasoning

145945.7%54.8%67.8%43.2%24.0%117910.9%89.1%60.280.0%

More about Arena Text Factuality

152339.7102552.0%47.0%47.8146987.1%49.5%76.7%63.7%26.5%150879.720.6%63.1%1.77×

More about Arena Coding

1466148458.0%1461147481.385.3%145462.5

More about Arena Writing, Literature & Language

30.1%146386.6%86.1%91.391.3%90.434.4%4.9%36.0%1487147480.120.7%4.2%80.0%37.6%62.0%11.9%14.0%22.0%89.9%44.5

By effortNone 13.2%Thinking 30.1%

More about Humanity's Last Exam

28.3%Reasoning effort: Budget 64K68.7%40.0%75.0%70.6%

More about Furniture Assembly

Specialized fields

1449143990.8%149815031487148382.0%50.6%43.5%14.3%62.1%39.2%30.6%73.2%39.2%91.0%77.5%78.9%76.0%94.0%86.5%94.1%83.4%30291.0%86.0%93.3%61.0%22.0%82.0%89.6%95.0%79.0%80.0%93.0%65.0%42.3%80.6%88.3%77.3%76.3%39.0%9.0%

More about Arena Korean

OddBench

90.0%Reasoning effort: High52.5%30

By effortNone 79.0%High 90.0%

More about BullshitBench v2

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