Skip to content

OpenAI

o1

OpenAI's first o-series model, trained to reason before it answers. Previews (o1-preview, o1-mini) on 12 September 2024; in the API from 17 December.

Announced
Released
Access
Public

Benchmark scores

TestScoreBest on the trackerHow it was measured
Epoch Capabilities Indexmany tests combined into one score141 pointsGPT-6.2 Astra 174 pointsIndependentlisted as "o1", CI 139.51 to 142.77Run 9 Nov 2026
Humanity's Last Examexpert questions, no tools8%GPT-6 Astra 55%Independentleaderboard entry "o1 (December 2024)" (o1-2024-12-17_unknown)
FrontierMath T1–3hard maths problems15%Claude Fable 5.2 95%Independenthigh reasoning effort (o1-2024-12-17_high)Run 28 Aug 2026
GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated77%Gemini 4 Argon 96%Independenthigh reasoning effort (o1-2024-12-17_high)Run 13 Feb 2025
OTIS Mock AIMEcompetition mathsSaturated73%GPT-5.5 100%Independentmedium reasoning effort (o1-2024-12-17_medium)Run 27 Feb 2025
METR time horizon (TH1.1 + TH2)task length an AI agent finishes half the time35 minGPT-6 Astra 1.7 daysIndependentPooled TH1.1 + TH2 suites; 95% CI 15.7–66.1 min
METR time horizontask length an AI agent finishes half the timeNo longer updated39 minClaude Mythos Preview 17 hIndependentMETR Horizon v1.1 run 'o1_inspect', medium reasoning effort (per Epoch AI mapping); 95% CI 21.2–65.0 min

Compare every model on the same tests →

Also on these charts

Sources