Frontier Log

AnthropicModel

Claude Sonnet 4.6

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

Benchmark results

BenchmarkScoreSource check
Humanity's Last ExamNo tools33.2%
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
re-graded with an updated grader model · reported later, in Introducing Claude Sonnet 534.6%
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
Humanity's Last ExamWith tools · web search, web fetch, code execution, programmatic tool calling; compaction at 50K up to 3M tokens; max effort49.0%
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
re-graded with an updated grader model · reported later, in Introducing Claude Sonnet 546.8%
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
SWE-bench Proreported later, in Introducing Claude Sonnet 5 · The source doesn't say whether this is the public or the commercial set58.1%
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 Diamond89.9%
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
BrowseCompweb search, web fetch, programmatic tool calling; compaction at 50K up to 10M tokens; max effort, no thinking74.7%
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
card update of 6 Mar 2026 after improved cheating detection; no thinking, max effort · in Claude Sonnet 4.6 System Card74.01%
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.
reported later, in Claude Sonnet 5 System Card76.2%
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.
ARC-AGI-2max effort, 120K thinking budget58.3%
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
high effort, 120K thinking budget; run by the ARC Prize Foundation on its private set60.4%
CorroboratedThe same number is printed in 1 other lab document: 1
Sources

Reported by Anthropic

Events

Stories

All models