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

BenchmarkScoreMetricPercentile
ATRBench23.7TSAcc Default (%) (self-reported)100
RAI-Bench - RAG Robustness (HY Abstention)91Rate (%)97.1
IMO-AnswerBench91.1Score (self-reported)96.4
Vellum - LiveCodeBench91.6Pass@1 (%)95.5
MERA - CheGeKa58.16F1 (%)95
MERA - USE67.55Grade, normalized (%)95
MERA Code - UnitTests31.95CodeBLEU (%)93.8
MERA Code - RealCodeJava37.25pass@1 (%)92.3
RAI-Bench - RAG Robustness (LC Abstention)93Rate (%)92
AI for Education Pedagogy - Social studies87.27Accuracy (%)91.4
Beyond the All-in-One Agent57.33avg. (self-reported)90.9
MERA Code0.47Total score90.8

Interactive version: theaggregate.ai/model?slug=deepseek-v4-flash · How It Works · Data refreshed daily, snapshot 2026-09-05.