MOONSHOT · MODEL TELEMETRY CARD

Kimi K3

Canonical UID: moonshot/kimi-k3 · Released: 2026-07-16

Measured configurations of Kimi K3 across benchmark suites. Exact machine values in site.json and data dumps.
Domain Benchmark Effort / Mode ⚡ Value/Cost # 🏆 Performance # Score Task $ Unit Cost $ Unit
Code ALE-Bench MAX #17 / 94 #19 / 94 1524.47 $1.28155 $0.0008407 rating point
Reason ProofBench STD #21 / 62 #10 / 62 0.87 $1.6468 $1.893 correct answer
Agent DeepSWE MAX #31 / 65 #18 / 65 0.685 $4.65468 $6.794 solved task

Compare against all models on ← #frontier Live Ranking Board · Fixed-capability cost half-life analysis on /methodology.