Kimi K3 (High): benchmark results
Provider: Moonshot. Released 2026-07-16. Access: Open.
Aggregate Rating 1812 ± 11, rank #120 of 2170 rated models, from 130 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| PPTBench | 67.8 | Mean reconstruction score (0-100): per task, the product of | 100 |
| Epoch AI - GPQA Diamond | 91.92 | Accuracy (%) | 91 |
| FrontierCS 2 - Approximate-MIPS Query Stopping | 86.13 | Task score (0–100) | 88.2 |
| FrontierCS 2 - GPU K-Means Kernel Optimization | 52.92 | Task score (0–100) | 88.2 |
| OckBench | 87.5 | Accuracy (%) | 86.4 |
| SelfBench - supabase/supabase | 77.78 | Task success (%) on 36 tasks rebuilt from the repository's m | 86.4 |
| SelfBench - vercel/vercel | 78.72 | Task success (%) on 47 tasks rebuilt from the repository's m | 85.7 |
| FrontierCS 2 - SQL Counterexample Search | 97.02 | Task score (0–100) | 85.3 |
| LLMConfBench | 0.38 | Conformer Ranking Accuracy (mean reliable-pair Kendall tau) | 85.2 |
| DSV-Mem | 24.3 | Accuracy (%), 1,000 stateful visual-memory questions, mean o | 84.6 |
| GameLogicBench - Combo | 42.86 | Solve rate (%; the 28 Combo tasks (several interacting mecha | 83.3 |
| Deep20Bench | 12.74 | Question score: subject-mean counted questions including mod | 82.4 |
Interactive version: theaggregate.ai/model?slug=kimi-k3-high · How It Works · Data refreshed daily, snapshot 2026-11-05.