Claude Opus 5 vs Kimi K3
Claude Opus 5 
Kimi K3 Verdict
- Claude Opus 5 leads on quality: 70.0 vs 66.5.
- Claude Opus 5 is stronger in agents & tools, coding, composite, human preference, knowledge, multimodal, reasoning.
- Kimi K3 is stronger in long context, maths.
- Kimi K3 is 1.7× cheaper ($6.00 vs $10.00 per 1M blended).
- Claude Opus 5 streams 1.4× faster (52 vs 37 tokens per second).
| Metric | Claude Opus 5 | Kimi K3 |
|---|---|---|
| BenchLeader Index | 70.0 | 66.5 |
| Agents & tools score | 71.6 | 67.5 |
| Coding score | 74.4 | 64.9 |
| Composite score | 92.9 | 74.1 |
| Human preference score | 78.2 | 71.2 |
| Knowledge score | 70.5 | 68.1 |
| Long context score | 66.2 | 71.1 |
| Maths score | 67.1 | 78.1 |
| Multimodal score | 69.4 | 65.1 |
| Reasoning score | 71.1 | – |
| Blended price $/M | $10.00 | $6.00 |
| Output speed | 52 tok/s | 37 tok/s |
| Time to first answer | 81.6 s | 57.7 s |
| Context window | 1M | 1.0M |
| GPQA Diamond | 92.9% | – |
| OTIS Mock AIME | 97.8% | – |
| SimpleBench | 80.6% | – |
| SciCode | – | 58.7% |
| WeirdML | 86.3% | – |
| APEX-Agents | – | 39.3% |
| FrontierCode | – | 44.2% |
| ProofBench | – | 87.0% |
| Epoch Capabilities Index | 162.6 | 157.6 |
| LiveBench | – | 79.2% |
| LiveBench Reasoning | – | 90.7% |
| LiveBench Coding | – | 81.5% |
| LiveBench Agentic Coding | – | 62.2% |
| LiveBench Mathematics | – | 84.4% |
| LiveBench Data Analysis | – | 78.7% |
| LiveBench Language | – | 85.5% |
| AA Intelligence Index | 50.7 | 43.8 |
| AA-LCR | 79.3% | 88.7% |
| MMMU-Pro | 84.7% | 80.5% |
| AA-Omniscience | 37.1 | 19.7 |
| GPQA Diamond (AA) | 93.2% | 93.5% |
| Humanity's Last Exam (AA) | 54.9% | 46.9% |
| SciCode (AA) | 56.4% | 59.5% |
| LiveCodeBench | 89.0% | – |
| MMLU-Pro | 91.6% | – |
| IOI | 91.7% | – |
| LegalBench | 87.0% | 86.0% |
| CorpFin | 73.2% | 71.6% |
| TaxEval | 75.1% | 75.7% |
| Terminal-Bench 2.1 (Vals) | 84.6% | 80.9% |
| SWE-bench (Vals) | 97.0% | 93.4% |
| GPQA Diamond (Vals) | 93.4% | 92.9% |
| Vals Index | 67.2 | 57.8 |
| HiL-Bench | 57.0% | – |
| EQ-Bench 4 | 1385 | 1339 |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.