Skip to content

Moonshot

Kimi K3.1 Pro

Kimi K3 with further training; 94.2% in Epoch AI's GPQA Diamond runs. The weights were promised by 4 November 2026, and the Hugging Face repository is live.

Announced
Released
Access
Public

Benchmark scores

TestScoreBest on the trackerHow it was measured
Epoch Capabilities Indexmany tests combined into one score162 pointsGPT-6.2 Astra 174 pointsIndependentlisted as "Kimi K3.1 Pro", CI 159.45 to 164.95Run 9 Nov 2026
AA Intelligence Index v5about ten tests in one score46.5 pointsClaude Fable 5.2 64.6 pointsIndependentv5.0.1; listed as Kimi K3.1 ProRun 10 Nov 2026
APEX-Agentsprofessional office tasks66%Gemini 4 Carbon 84%Independentpass@1, unknown effort; listed as "Kimi K3.1 Pro"
FrontierMath T1–3hard maths problems78%Claude Fable 5.2 95%Independentmax reasoning effort, via Together AI (kimi-k3.1-pro-together_max)Run 6 Nov 2026
GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated94%Gemini 4 Argon 96%Independentmax reasoning effort, via Together AI (kimi-k3.1-pro-together_max)Run 6 Nov 2026
FrontierMath T4research-level maths problemsSaturated66%GPT-6.1 Sol 100%Independentmax reasoning effort, via Together AI (kimi-k3.1-pro-together_max)Run 6 Nov 2026
OTIS Mock AIMEcompetition mathsSaturated100%Top scoreIndependentmax reasoning effort, via Together AI (kimi-k3.1-pro-together_max)Run 6 Nov 2026

Compare every model on the same tests →

Sources