DeepSeek V4 Flash (0731) (Non-reasoning): benchmark results
Provider: DeepSeek. Released 2026-07-31. Access: Open.
Aggregate Rating 1566 ± 26, rank #846 of 2201 rated models, from 20 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| World Model Situational Probe Passing - Coverage | 0.38 | Feature-multiset structural agreement among probe-passing ca | 75 |
| LLM Milgram Obedience - Baseline Mean Breakoff Voltage | 242.14 | Mean last shock administered in valid baseline sessions (vol | 63.2 |
| LLM Milgram Obedience - Baseline Full-Obedience Rate | 21.43 | Valid baseline sessions ending at 450 V (%; lower is less ob | 60.5 |
| World Model Situational Probe Passing - F1 | 0.48 | Feature-multiset structural agreement among probe-passing ca | 50 |
| World Model Situational Probe Passing - Precision | 0.82 | Feature-multiset structural agreement among probe-passing ca | 50 |
| World Model POSG Game - Coverage | 0.33 | Permutation-invariant structural agreement (0-1) | 25 |
| World Model POSG Game - F1 | 0.27 | Permutation-invariant structural agreement (0-1) | 25 |
| World Model POSG Game - Precision | 0.27 | Permutation-invariant structural agreement (0-1) | 25 |
| ObviousBench | 62.5 | Answer pass³ (%) | 18.8 |
| Chess Bench LLM | -189 | Lichess-Classical-calibrated Glicko-2 rating (native signed | 12.1 |
| FlavourBench | 55.43 | FlavourBench Score (0-100) | 11.5 |
| FlavourBench - Substitution | 63.06 | Substitution Family Score (0-100) | 11.5 |
Interactive version: theaggregate.ai/model?slug=deepseek-v4-flash-0731-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-12.