Skip to content

Anthropic

Claude Opus 4.6

First Opus to take prompts of 1 million tokens (in beta). METR time horizon: about 12 hours (range about 5 to 60).

Announced
Access
Public

Benchmark scores

TestScoreBest on the trackerHow it was measured
Epoch Capabilities Indexmany tests combined into one score156 pointsGPT-6.2 Astra 174 pointsIndependentlisted as "Claude Opus 4.6", CI 154.57 to 158.22Run 9 Nov 2026
Remote Labor Indexpaid freelance projects4.2%Claude Fable 5.2 41%Independentclaude-opus-4-6 (CoWork)Run 5 Mar 2026
APEX-Agentsprofessional office tasks46%Gemini 4 Carbon 84%Independentpass@1, max effort; listed as "Opus 4.6"
Humanity's Last Examexpert questions, no tools34%GPT-6 Astra 55%Independentleaderboard entry "claude-opus-4-6-thinking-max" (claude-opus-4-6_max)
ARC-AGI-3puzzle games, standard harnessTypical first-time human player: 100%0.5%Claude Fable 5.2 81%IndependentARC Prize config "Anthropic Opus 4.6 (Max)" (anthropic-opus-4-6-max-effort), standard harness
FrontierMath T1–3hard maths problems66%Claude Fable 5.2 95%Independentmax reasoning effort (claude-opus-4-6_max)Run 11 Jun 2026
GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated91%Gemini 4 Argon 96%Independent32K thinking budget (claude-opus-4-6_32K)Run 6 Feb 2026
FrontierMath T4research-level maths problemsSaturated27%GPT-6.1 Sol 100%Independentmax reasoning effort (claude-opus-4-6_max)Run 11 Jun 2026
ARC-AGI-2visual puzzles, model aloneAverage person: 60%Saturated69%GPT-6 Astra 95%IndependentARC Prize config "Claude Opus 4.6 (120K, High)" (claude-opus-4-6-thinking-120K-high)
OTIS Mock AIMEcompetition mathsSaturated94%GPT-5.5 100%Independent64K thinking budget (claude-opus-4-6_64K)Run 6 Feb 2026
SWE-bench Verifiedfixing real software bugsKnown flaws79%Claude Opus 4.7 83%Independentdefault settings (claude-opus-4-6)Run 18 Feb 2026
METR time horizon (TH1.1 + TH2)task length an AI agent finishes half the time6.4 hGPT-6 Astra 1.7 daysIndependentPooled TH1.1 + TH2 suites; 95% CI 193.4–729.6 min
METR time horizontask length an AI agent finishes half the timeNo longer updated12 hClaude Mythos Preview 17 hIndependentMETR Horizon v1.1 run 'claude_opus_4_6_inspect'; 95% CI 316.7–3633.8 min

Compare every model on the same tests →

Also on these charts

Sources