Kimi K3 scores 50.6 on the Respan Index (#21), ahead of DeepSeek V4 Flash at 43.3 (#37). Kimi K3 leads on 18 of the 23 benchmarks both report. Kimi K3 has the larger context window (1.0M tokens).
* Estimated: no published scores in that category. How the Respan Index works
| 91% |
| 93.5% |
| LiveBench Agentic Coding (LiveBench) | 46.8% | 62.2% |
| LiveBench Coding (LiveBench) | 75% | 81.4% |
| LiveBench Data Analysis (LiveBench) | 79.3% | 78.7% |
| LiveBench Instruction Following (LiveBench) | 65.5% | 71.4% |
| LiveBench Language (LiveBench) | 79.2% | 85.5% |
| LiveBench Mathematics (LiveBench) | 86.8% | 84.4% |
| LiveBench Reasoning (LiveBench) | 86.6% | 90.7% |
| Mystery Game Puzzles (Epoch AI) | 34% | 26% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 94.4% | 97.2% |
| SimpleQA Verified | 33.6% | 50.6% |
| Terminal-Bench 4.0 (Vals AI) | 18.7% | 17.2% |
| WeirdML (Håvard Tveit Ihle) | 63% | 82.6% |
| LMArena Elo | 1436 | 1488 |
| LMArena WebDev (LMArena) | 1581 | 1658 |
| SimpleBench (SimpleBench) | 61.1% | 60.7% |
| Agents' Last Exam | 25.2% | 28.3% |
| Toolathlon-Verified | 70.3% | 76.5% |
| Toolathlon-Verified (HKUST) | 70.7% | 76.5% |
Higher is better. Each score links to where it was published.