Moonshot AIopen weightsCurrentReasoning
Kimi K3
2.8T MoE, 896 experts; weights released Jul 27, 2026 under a custom licence with revenue-share above $20M/yr. The strongest open model on most boards.
Jul 16, 2026
$3 / $15
$6
1M
Benchmark scores
| Benchmark | Score | Δ vs prior | Rank (current) | Configuration | Source |
|---|---|---|---|---|---|
Terminal-Bench 2.1 | 80.9% | #6 of 22 | leaderboard | benchlm.ai ↗ | |
SWE-bench Verified | 93.4% | #1 of 8 | Vals AI run · independent run | morphllm.com ↗ | |
GPQA Diamond | 93.5% | #5 of 21 | leaderboard | benchlm.ai ↗ | |
Humanity's Last Exam | 43.5% | #14 of 31 | no tools; Moonshot reports 56.0 with tools · lab-reported | kimi.com ↗ | |
ARC-AGI-2 | 60.4% | #8 of 9 | leaderboard | benchlm.ai ↗ | |
MMLU-Pro | 88% | #4 of 12 | Vals AI run, max reasoning · independent run | vals.ai ↗ | |
Artificial Analysis Intelligence Index | 57.1 | #2 of 16 | leaderboard | benchlm.ai ↗ | |
LMArena Text | 1476 | #3 of 4 | max · leaderboard | benchlm.ai ↗ |