DeepSeek V4 Flash: benchmark results
DeepSeek's efficient open V4 Flash MoE model (13B active). Provider: DeepSeek. Released 2026-04-23. Access: Open.
Unified ELO 1640 ± 1, rank #147 of 1392 rated models, from 304 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ATRBench | 23.7 | TSAcc Default (%) (self-reported) | 100 |
| RAI-Bench - RAG Robustness (HY Abstention) | 91 | Rate (%) | 97.1 |
| IMO-AnswerBench | 91.1 | Score (self-reported) | 96.4 |
| Vellum - LiveCodeBench | 91.6 | Pass@1 (%) | 95.5 |
| MERA - CheGeKa | 58.16 | F1 (%) | 95 |
| MERA - USE | 67.55 | Grade, normalized (%) | 95 |
| MERA Code - UnitTests | 31.95 | CodeBLEU (%) | 93.8 |
| MERA Code - RealCodeJava | 37.25 | pass@1 (%) | 92.3 |
| RAI-Bench - RAG Robustness (LC Abstention) | 93 | Rate (%) | 92 |
| AI for Education Pedagogy - Social studies | 87.27 | Accuracy (%) | 91.4 |
| Beyond the All-in-One Agent | 57.33 | avg. (self-reported) | 90.9 |
| MERA Code | 0.47 | Total score | 90.8 |
Interactive version: theaggregate.ai/model?slug=deepseek-v4-flash · How It Works · Data refreshed daily, snapshot 2026-09-05.