Kimi K3 vs Qwen3 8
Kimi K3 
Qwen3 8 Verdict
- Kimi K3 leads on quality: 66.5 vs 63.9.
- Kimi K3 is stronger in agents & tools, composite, human preference, knowledge, long context, maths.
- Qwen3 8 (max) is stronger in coding, multimodal, reasoning.
- Qwen3 8 (max) is 2.3× cheaper ($2.67 vs $6.00 per 1M blended).
| Metric | Kimi K3 | Qwen3 8 (max) |
|---|---|---|
| BenchLeader Index | 66.5 | 63.9 |
| Agents & tools score | 67.5 | 56.1 |
| Coding score | 64.9 | 71.7 |
| Composite score | 74.1 | 70.8 |
| Human preference score | 71.2 | 67.9 |
| Knowledge score | 68.1 | 62.3 |
| Long context score | 71.1 | 65.7 |
| Maths score | 78.1 | – |
| Multimodal score | 65.1 | 67.2 |
| Reasoning score | – | 67.7 |
| Blended price $/M | $6.00 | $2.67 |
| Output speed | 37 tok/s | 39 tok/s |
| Time to first answer | 57.7 s | 53.2 s |
| Context window | 1.0M | 1M |
| SciCode | 58.7% | – |
| APEX-Agents | 39.3% | – |
| FrontierCode | 44.2% | – |
| ProofBench | 87.0% | – |
| Epoch Capabilities Index | 157.6 | 156.6 |
| LMArena Text | – | 1480 |
| LMArena Hard Prompts | – | 1502 |
| LMArena Coding | – | 1520 |
| LMArena WebDev | – | 1685 |
| LMArena Vision | – | 1313 |
| LMArena Agent | – | 3.9 |
| LiveBench | 79.2% | 78.5% |
| LiveBench Reasoning | 90.7% | 88.2% |
| LiveBench Coding | 81.5% | 72.9% |
| LiveBench Agentic Coding | 62.2% | 64.7% |
| LiveBench Mathematics | 84.4% | 91.3% |
| LiveBench Data Analysis | 78.7% | 78.4% |
| LiveBench Language | 85.5% | 79.7% |
| AA Intelligence Index | 43.8 | 40.3 |
| AA-LCR | 88.7% | 78.3% |
| MMMU-Pro | 80.5% | 82.3% |
| AA-Omniscience | 19.7 | 3.4 |
| GPQA Diamond (AA) | 93.5% | 92.7% |
| Humanity's Last Exam (AA) | 46.9% | 43.0% |
| SciCode (AA) | 59.5% | 53.2% |
| LiveCodeBench | – | 87.8% |
| MMLU-Pro | – | 88.6% |
| IOI | – | 73.0% |
| LegalBench | 86.0% | 83.6% |
| CorpFin | 71.6% | 65.8% |
| TaxEval | 75.7% | 75.5% |
| Terminal-Bench 2.1 (Vals) | 80.9% | 67.4% |
| SWE-bench (Vals) | 93.4% | 85.6% |
| GPQA Diamond (Vals) | 92.9% | 93.7% |
| Vals Index | 57.8 | 51.8 |
| EQ-Bench 4 | 1339 | – |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.