Kimi K3 scores 50.6 on the Respan Index (#21), ahead of Grok 4.7 at 49.8 (#23). Kimi K3 leads on 14 of the 24 benchmarks both report. Grok 4.7 is 2.0x cheaper per token at list price. Kimi K3 has the larger context window (1.0M tokens).
* Estimated: no published scores in that category. How the Respan Index works
| FrontierSWE V2 (Proximal Labs) |
| 29.5% |
| 25.9% |
| Furniture Assembly (Epoch AI) | 20.8% | 34.2% |
| GPQA Diamond | 92.7% | 93.5% |
| LiveBench Agentic Coding (LiveBench) | 54% | 62.2% |
| LiveBench Coding (LiveBench) | 77.2% | 81.4% |
| LiveBench Data Analysis (LiveBench) | 76.9% | 78.7% |
| LiveBench Instruction Following (LiveBench) | 75.3% | 71.4% |
| LiveBench Language (LiveBench) | 80.1% | 85.5% |
| LiveBench Mathematics (LiveBench) | 95.7% | 84.4% |
| LiveBench Reasoning (LiveBench) | 82.7% | 90.7% |
| Mystery Game Puzzles (Epoch AI) | 29% | 26% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 98.1% | 97.2% |
| SAGE (Vals AI) | 40.8% | 52.8% |
| SimpleQA Verified | 56% | 50.6% |
| Terminal-Bench 4.0 (Vals AI) | 28.8% | 17.2% |
| Vending-Bench 2 (Andon Labs) | 10536.83 | 5165.04 |
| LMArena Agent (LMArena) | 0.0404 | 0.0418 |
| LMArena Elo | 1443 | 1488 |
| LMArena WebDev (LMArena) | 1638 | 1658 |
| DeepSWE v1.1 | 71% | 67.3% |
Higher is better. Each score links to where it was published.