DeepSeek V4 Flash (0731) — benchmark results
Provider: DeepSeek. Released 2026-07-31. Access: Open.
Unified ELO 1861 ± 18, rank #101 of 1839 rated models, from 38 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AI Chess Leaderboard (Reasoning) | 1645 | Elo | 96.7 |
| LLM Stats Score | 48.49 | LLM Stats Score (conservative rating) | 96.2 |
| PM-LLM-Benchmark | 37.7 | Score | 95.8 |
| AI Chess Leaderboard (Continuation) | 1409 | Elo | 94 |
| Vals AI SWE-bench Verified | 88.8 | Resolved (%) | 93.4 |
| LLM Stats (Toolathlon) | 70.3 | Score (%) | 93.3 |
| Vals AI LiveCodeBench | 87.26 | Accuracy (%) | 92.5 |
| LLM Stats (NL2Repo) | 54.2 | Score (%) | 92.3 |
| Vals AI Vibe Code Bench | 74.74 | Accuracy (%) | 90.9 |
| Vals AI GPQA | 89.9 | Accuracy (%) | 84.4 |
| Vals AI Harvey Legal Agent Bench | 8.33 | Accuracy (%) | 82.8 |
| Vals AI CyberBench | 77.45 | Accuracy (%) | 81 |
Interactive version: theaggregate.ai/model?slug=deepseek-v4-flash-0731 · How It Works · Data refreshed daily, snapshot 2026-08-05.