DeepSeek V4 Flash (Reasoning): benchmark results

Provider: DeepSeek. Released 2026-04-23. Access: Open.

Unified ELO 1632 ± 1, rank #310 of 1761 rated models, from 9 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
UGI Leaderboard56.63UGI Score97.4
UGI - Natural Intelligence45.52NatInt Score88.4
UGI - Writing49.74Writing Score88.2
Wolfram LLM Benchmarking Project59.6Correct Functionality (%)88
AtmosCoder-Bench91.1Accuracy (%, mean of 3 runs)66.7
GRIPS88.2Accuracy (%)61.5
GRIPS-hard49.3Accuracy (%)61.5
UGI - Willingness (W/10)5.2W/10 Score42
SpeechMap Compliance50.7% Requests Completed33.5

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