DeepSeek V4 Flash (Max): benchmark results
DeepSeek V4 Flash evaluated at the max reasoning-effort setting. Provider: DeepSeek. Released 2026-04-24. Access: Open.
Aggregate Rating 1727 ± 8, rank #288 of 2168 rated models, from 227 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SuperCLUE General (May 2026) - Math Reasoning | 82.69 | Score | 100 |
| Upstage Solar Open 2 Card - HRM8K | 93.4 | Accuracy (%) | 100 |
| Upstage Solar Open 2 Card - KMMLU-Pro | 78.9 | Accuracy (%) | 100 |
| Upstage Solar Open 2 Card - Ko-AIME 2025 | 98 | Accuracy (%) | 100 |
| tau-Rec | 57.1 | pass^1 (0-1 scaled to %), the mean strict success rate per t | 100 |
| tau-Rec - Constraint Score | 59.6 | Mean constraint sub-score (0-1 scaled to %), the partial-cre | 100 |
| Nejumi 4 - jaster (0-shot) - JSICK | 85 | Exact match (%) | 96.2 |
| SuperCLUE General (March 2026) - Math Reasoning | 89.08 | Score | 95.7 |
| Nejumi 4 - jaster (0-shot) - JBLiMP | 97 | Exact match (%) | 95.5 |
| AGI-Eval Community - Mathematical Reasoning | 86.96 | Accuracy (%) | 94.3 |
| AGI-Eval Community - Subject Reasoning (Chinese) | 92.88 | Accuracy (%) | 93.8 |
| AGI-Eval Community - Algorithmic Reasoning | 68.42 | Accuracy (%) | 92.1 |
Interactive version: theaggregate.ai/model?slug=deepseek-v4-flash-max · How It Works · Data refreshed daily, snapshot 2026-10-20.