Grok 4.5 vs Kimi K3
Grok 4.5 
Kimi K3 Verdict
- Kimi K3 leads on quality: 66.5 vs 63.4.
- Grok 4.5 is stronger in knowledge, reasoning.
- Kimi K3 is stronger in agents & tools, coding, composite, human preference, long context, multimodal, maths.
- Grok 4.5 is 2.0× cheaper ($3.00 vs $6.00 per 1M blended).
- Grok 4.5 streams 1.5× faster (56 vs 37 tokens per second).
| Metric | Grok 4.5 | Kimi K3 |
|---|---|---|
| BenchLeader Index | 63.4 | 66.5 |
| Agents & tools score | 58.6 | 67.5 |
| Coding score | 59.9 | 64.9 |
| Composite score | 66.0 | 74.1 |
| Human preference score | 66.9 | 71.2 |
| Knowledge score | 76.0 | 68.1 |
| Long context score | 66.2 | 71.1 |
| Multimodal score | 64.8 | 65.1 |
| Reasoning score | 68.9 | – |
| Maths score | – | 78.1 |
| Blended price $/M | $3.00 | $6.00 |
| Output speed | 56 tok/s | 37 tok/s |
| Time to first answer | 12.7 s | 57.7 s |
| Context window | 500k | 1.0M |
| SimpleBench | 70.0% | – |
| SciCode | – | 58.7% |
| WeirdML | 46.4% | – |
| APEX-Agents | 34.2% | 39.3% |
| FrontierCode | 42.4% | 44.2% |
| ProofBench | – | 87.0% |
| Epoch Capabilities Index | 153.9 | 157.6 |
| LMArena Text | 1471 | – |
| LMArena Hard Prompts | 1495 | – |
| LMArena Coding | 1523 | – |
| LMArena WebDev | 1556 | – |
| LMArena Vision | 1291 | – |
| LMArena Agent | 3.9 | – |
| LiveBench | 75.8% | 79.2% |
| LiveBench Reasoning | 87.2% | 90.7% |
| LiveBench Coding | 68.6% | 81.5% |
| LiveBench Agentic Coding | 56.5% | 62.2% |
| LiveBench Mathematics | 90.8% | 84.4% |
| LiveBench Data Analysis | 73.0% | 78.7% |
| LiveBench Language | 82.8% | 85.5% |
| AA Intelligence Index | 39.1 | 43.8 |
| AA-LCR | 79.3% | 88.7% |
| MMMU-Pro | 80.4% | 80.5% |
| AA-Omniscience | 25.3 | 19.7 |
| GPQA Diamond (AA) | 93.1% | 93.5% |
| Humanity's Last Exam (AA) | 42.7% | 46.9% |
| SciCode (AA) | 55.0% | 59.5% |
| 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 |
| EQ-Bench 4 | – | 1339 |
| Kagi LLM Benchmark | 83.5% | – |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.