GPT-5.6 Sol vs Kimi K3
GPT-5.6 Sol 
Kimi K3 Verdict
- GPT-5.6 Sol (high) leads on quality: 68.0 vs 66.5.
- GPT-5.6 Sol (high) is stronger in agents & tools, coding, composite, instruction following, knowledge, multimodal, reasoning.
- Kimi K3 is stronger in long context, human preference, maths.
- Kimi K3 is 1.3× cheaper ($6.00 vs $8.00 per 1M blended).
- GPT-5.6 Sol (high) streams 1.6× faster (59 vs 37 tokens per second).
| Metric | GPT-5.6 Sol (high) | Kimi K3 |
|---|---|---|
| BenchLeader Index | 68.0 | 66.5 |
| Agents & tools score | 87.6 | 67.5 |
| Coding score | 72.0 | 64.9 |
| Composite score | 82.5 | 74.1 |
| Instruction following score | 67.6 | – |
| Knowledge score | 73.7 | 68.1 |
| Long context score | 67.4 | 71.1 |
| Multimodal score | 66.5 | 65.1 |
| Reasoning score | 62.9 | – |
| Human preference score | – | 71.2 |
| Maths score | – | 78.1 |
| Blended price $/M | $8.00 | $6.00 |
| Output speed | 59 tok/s | 37 tok/s |
| Time to first answer | 31.4 s | 57.7 s |
| Context window | 1M | 1.0M |
| SciCode | 56.9% | 58.7% |
| WeirdML | 88.8% | – |
| APEX-Agents | – | 39.3% |
| FrontierCode | – | 44.2% |
| ProofBench | – | 87.0% |
| Epoch Capabilities Index | – | 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 | 42.5 | 43.8 |
| IFBench | 69.2% | – |
| AA-LCR | 81.7% | 88.7% |
| MMMU-Pro | 81.8% | 80.5% |
| AA-Omniscience | 20.4 | 19.7 |
| Terminal-Bench Hard | 62.1% | – |
| GPQA Diamond (AA) | 92.8% | 93.5% |
| Humanity's Last Exam (AA) | 46.0% | 46.9% |
| SciCode (AA) | 57.8% | 59.5% |
| τ²-Bench Telecom (AA) | 83.3% | – |
| LegalBench | – | 86.0% |
| CorpFin | – | 71.6% |
| TaxEval | – | 75.7% |
| Terminal-Bench 2.1 (Vals) | – | 80.9% |
| SWE-bench (Vals) | – | 93.4% |
| GPQA Diamond (Vals) | – | 92.9% |
| Vals Index | – | 57.8 |
| EnigmaEval | 37.1% | – |
| EQ-Bench 4 | – | 1339 |
| ARC-AGI-1 | 97.0% | – |
| ARC-AGI-2 | 85.4% | – |
| ARC-AGI-3 | 2.2% | – |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.