Frontier Log

AnthropicModel

Claude Opus 4.6

Made by
Anthropic
Released
Announced in
Claude Opus 4.6anthropic.com
Event
Claude Opus 4.6
API record
/api/v1/models/claude-opus-4-6

Benchmark results

BenchmarkScoreSource check
Humanity's Last ExamNo tools40.0%
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 6 other lab documents, the first 3 linked: 1 2 3
Humanity's Last ExamWith tools53.0%
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.
CorroboratedThe same number is printed in 2 other lab documents: 1 2
reported later, in Introducing Claude Opus 4.753.3%
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
SWE-bench Verifiedaveraged over 25 trials80.8%
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
SWE-bench Promemorization screens flag a subset of problems · reported later, in Introducing Claude Opus 4.7 · The source doesn't say whether this is the public or the commercial set53.4%
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
GPQA Diamond91.3%
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
BrowseCompweb search, web fetch, programmatic tool calling; compaction at 50K up to 10M tokens; max effort, no thinking84.0%
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.
CorroboratedThe same number is printed in 3 other lab documents: 1 2 3
card update of 6 Mar 2026 after improved cheating detection; no thinking, max effort · in Claude Opus 4.6 System Card83.7%
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 4 other lab documents, the first 3 linked: 1 2 3
ARC-AGI-2max effort, 120K thinking budget68.8%
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 3 other lab documents: 1 2 3
ARC Prize Foundation run: high effort, 120K thinking tokens, private set · in Claude Opus 4.6 System Card69.17%
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.
Terminal-Bench 2.0Terminus 2 harness; 5 to 15 samples per task65.4%
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 8 other lab documents, the first 3 linked: 1 2 3
τ²-benchTelecom99.3%
CheckedA person checked this number against its source on 18 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
Sources

Reported by Anthropic

Events

Stories

All models