Skip to content

Anthropic

Claude Opus 4.5

Lets API users choose how much effort the model spends. METR time horizon: about 4.9 hours.

Announced
Access
Public

Benchmark scores

TestScoreBest on the trackerHow it was measured
Epoch Capabilities Indexmany tests combined into one score152 pointsGPT-6.2 Astra 174 pointsIndependentlisted as "Claude Opus 4.5", CI 150.28 to 153.47Run 9 Nov 2026
Remote Labor Indexpaid freelance projects3.8%Claude Fable 5.2 41%Independentclaude-opus-4-5-20251101-thinkingRun 9 Jan 2026
Humanity's Last Examexpert questions, no tools25%GPT-6 Astra 55%Independentleaderboard entry "claude-opus-4-5-20251101-thinking" (claude-opus-4-5-20251101_unknown)
FrontierMath T1–3hard maths problems34%Claude Fable 5.2 95%Independent32K thinking budget (claude-opus-4-5-20251101_32K)Run 11 Jun 2026
GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated86%Gemini 4 Argon 96%Independent32K thinking budget (claude-opus-4-5-20251101_32K)Run 24 Nov 2025
FrontierMath T4research-level maths problemsSaturated4.9%GPT-6.1 Sol 100%Independent32K thinking budget (claude-opus-4-5-20251101_32K)Run 11 Jun 2026
ARC-AGI-2visual puzzles, model aloneAverage person: 60%Saturated38%GPT-6 Astra 95%IndependentARC Prize config "Opus 4.5 (Thinking, 64K)" (claude-opus-4-5-20251101-thinking-64k)
OTIS Mock AIMEcompetition mathsSaturated86%GPT-5.5 100%Independent32K thinking budget (claude-opus-4-5-20251101_32K)Run 24 Nov 2025
SWE-bench Verifiedfixing real software bugsKnown flaws77%Claude Opus 4.7 83%Independentdefault settings (claude-opus-4-5-20251101)Run 5 Feb 2026
METR time horizon (TH1.1 + TH2)task length an AI agent finishes half the time4.6 hGPT-6 Astra 1.7 daysIndependentPooled TH1.1 + TH2 suites; 95% CI 144.7–511.0 min
METR time horizontask length an AI agent finishes half the timeNo longer updated4.9 hClaude Mythos Preview 17 hIndependentMETR Horizon v1.1 run 'claude_opus_4_5_inspect'; 95% CI 161.7–623.7 min

Compare every model on the same tests →

Also on these charts

Sources