Moonshot · model passport
Kimi K2 Preview
kimi-k2-preview- Open weights
- Modified MIT licence
- Released 2026-10-28(first seen on LMArena (approximate))
- Overall · LMArena
- 1,418#137 of 469
- Best percentile
- 72.1Coding
- Cohort standing
- 55.7median 57.4
What the data says
Benchmarked on 5 of 27 tracked axes. Every figure below is the model's own published result on that board.
Uneven: 54.2 percentile points between its best and worst axis — 72.1 pct on Coding versus 17.9 pct on Factuality Arena.
- Coding72.1 pct1,468Verify →
- Overall71 pct1,418Verify →
- Hard prompts70.8 pct1,438Verify →
- Instruction following64 pct1,390Verify →
- Factuality Arena17.9 pct1,402Verify →
- WebDev ArenaNot benchmarked
- Vision ArenaNot benchmarked
- Search ArenaNot benchmarked
- Document ArenaNot benchmarked
- Agent ArenaNot benchmarked
- GPQA DiamondNot benchmarked
- AIME (math)Not benchmarked
- FrontierMathNot benchmarked
- SimpleQA VerifiedNot benchmarked
- ARC-AGI-2Not benchmarked
- Humanity's Last ExamNot benchmarked
- SimpleBenchNot benchmarked
- LiveBenchNot benchmarked
- BFCL v4 (function calling)Not benchmarked
- SWE-bench Verified (live board)Not benchmarked
- Terminal-Bench 4.0Not benchmarked
- MMLU-Pro (Artificial Analysis)Not benchmarked
- LiveCodeBench (Artificial Analysis)Not benchmarked
- IFBench (Artificial Analysis)Not benchmarked
- HHEM Hallucination (Vectara)Not benchmarked
- SWE-bench VerifiedNot benchmarked
- Terminal-BenchNot benchmarked
Percentiles are the model's standing within each board's own population. Axes marked "not benchmarked" have no published result for this model — that is an absence of data, not a low score.
Kimi K2 Preview (0905) from moonshot covers 5 of 27 axes with a composite score of 59.2 and a shrunk score of 55.7, achieving percentile 55.7 on the standing metric.
standingVerify →In the 180-day cohort of 215 models, Kimi K2 Preview (0905) from moonshot achieved a standing of 55.7, which is below the cohort median standing of 57.4.
cohortVerify →
Kimi K2 Preview (0905) from moonshot shows a sensitivity spread of 403 between the Coding-focused and All sources templates across 5 covered templates, with ranks ranging from 100 to 503.
rank swingVerify →
Every claim is a computed fact from the current board data, phrased by a low-temperature model call and rejected by a validation gate if any number is not in the source row. How the leaderboards work →