DeepSeek V4 Flash scores 43.3 on the Respan Index (#37), ahead of Kimi K2.7 Code at 34.1 (#58). DeepSeek V4 Flash leads on 16 of the 18 benchmarks both report. DeepSeek V4 Flash has the larger context window (1M tokens).
* Estimated: no published scores in that category. How the Respan Index works
| GPQA Diamond |
| 91% |
| 87.9% |
| LiveBench Agentic Coding (LiveBench) | 46.8% | 45.7% |
| LiveBench Coding (LiveBench) | 75% | 74% |
| LiveBench Data Analysis (LiveBench) | 79.3% | 62.7% |
| LiveBench Instruction Following (LiveBench) | 65.5% | 56.3% |
| LiveBench Language (LiveBench) | 79.2% | 77.9% |
| LiveBench Mathematics (LiveBench) | 86.8% | 79.6% |
| LiveBench Reasoning (LiveBench) | 86.6% | 82.8% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 94.4% | 95.6% |
| SimpleQA Verified | 33.6% | 36.5% |
| WeirdML (Håvard Tveit Ihle) | 63% | 54.1% |
| LMArena WebDev (LMArena) | 1581 | 1473 |
| SimpleBench (SimpleBench) | 61.1% | 57.9% |
| Toolathlon-Verified (HKUST) | 70.7% | 58% |
Higher is better. Each score links to where it was published.