Frontier Log

AnthropicModel

Claude Sonnet 5.5

Made by
Anthropic
Released
Announced in
Introducing Claude Sonnet 5.5anthropic.com
Event
Claude Sonnet 5.5
API record
/api/v1/models/claude-sonnet-5-5

Benchmark results

BenchmarkScoreSource check
Humanity's Last ExamNo tools56.9%
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
Humanity's Last ExamWith tools64.5%
CheckedA person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
CorroboratedThe same number is printed in 3 other lab documents: 1 2 3
SWE-bench Proadaptive thinking, max effort; average of 5 trials · The source doesn't say whether this is the public or the commercial set81.3%
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
Terminal-Bench 4.0adaptive thinking70.6%
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 3 other lab documents: 1 2 3
Terminal-Bench Science 0.1Claude Code in --bare mode; max effort; standard error ±4.7 points59.9%
CheckedA person checked this number against its source on 19 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
AutomationBenchrun by Zapier; private set, release 1.0.6; max effort; default fallbacks on44.7%
CheckedA person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
OSWorld 2.1Full set, partial score · max effort; mean of 5 runs; September 10, 2026 task files80.1%
CorroboratedThe same number is printed in 1 other lab document: 1
OSWorld 2.1Full set, strict pass rate · max effort; mean of 5 runs; September 10, 2026 task files43.5%
CheckedA person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
OSWorld 2.1Offline set, partial score · 82-task offline subset, no internet access; max effort; mean of 5 attempts · reported later, in Introducing Claude Haiku 5.583.9%
CorroboratedThe same number is printed in 1 other lab document: 1
OSWorld 2.1Offline set, strict pass rate · 82-task offline subset, no internet access; max effort; mean of 5 attempts · reported later, in Claude Haiku 5.5 System Card48.8%
CheckedA person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
FrontierCode 1.1Main set · xhigh effort; run by Cognition in Claude Code52.1%
CorroboratedThe same number is printed in 3 other lab documents: 1 2 3
max effort; run by Cognition in Claude Code46.2%
CorroboratedThe same number is printed in 2 other lab documents: 1 2
FrontierCode 1.1Extended set · xhigh effort; run by Cognition in Claude Code64.4%
CorroboratedThe same number is printed in 1 other lab document: 1
max effort; run by Cognition in Claude Code59.1%
CorroboratedThe same number is printed in 1 other lab document: 1
DeepSWE v1.1average of 5 trials71.0%
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.
Sources

Reported by Anthropic

Events

Stories

All models