Frontier Log

AnthropicModel

Claude Sonnet 4.5

Made by
Anthropic
Released
Announced in
Introducing Claude Sonnet 4.5anthropic.com
Event
Claude Sonnet 4.5
API record
/api/v1/models/claude-sonnet-4-5

Benchmark results

BenchmarkScoreSource check
Humanity's Last ExamNo tools · decontamination improved with Opus 4.5, which may affect this score · reported later, in Claude Opus 4.5 System Card17.7%
CheckedA person checked this number against its source on 16 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
CorroboratedThe same number is printed in 1 other lab document: 1
Humanity's Last ExamWith tools · web search, web fetch and code execution, no reasoning; Claude Sonnet 4.5 as grader · reported later, in Claude Opus 4.5 System Card28.4%
CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
web search, web fetch, code execution, programmatic tool calling; compaction at 50K up to 3M tokens; max effort, adaptive thinking · reported later, in Claude Opus 4.633.6%
CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
CorroboratedThe same number is printed in 1 other lab document: 1
SWE-bench Verifiedbash and file-editing tools; 200K thinking budget; averaged over 10 trials; no test-time compute77.2%
CheckedA person checked this number against its source on 16 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
CorroboratedThe same number is printed in 14 other lab documents, the first 3 linked: 1 2 3
parallel test-time compute · parallel test-time compute82.0%
CheckedA person checked this number against its source on 16 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
GPQA Diamond83.4%
CheckedA person checked this number against its source on 16 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
CorroboratedThe same number is printed in 7 other lab documents, the first 3 linked: 1 2 3
BrowseCompreported later, in Claude Opus 4.643.9%
CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
CorroboratedThe same number is printed in 1 other lab document: 1
ARC-AGI-2reported later, in Introducing Claude Opus 4.513.6%
CheckedA person checked this number against its source on 16 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
CorroboratedThe same number is printed in 4 other lab documents, the first 3 linked: 1 2 3
Terminal-Bench 2.0reported later, in Introducing Claude Opus 4.550.0%
CheckedA person checked this number against its source on 16 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
CorroboratedThe same number is printed in 2 other lab documents: 1 2
reported later, in Claude Opus 4.6 System Card51.0%
CorroboratedThe same number is printed in 3 other lab documents: 1 2 3
Sources

Reported by Anthropic

Events

Stories

All models