DeepSeek-V4-Flash-FP8 — benchmark results

DeepSeek V4 Flash evaluated in FP8 quantization. Provider: DeepSeek. Released 2026-04-23. Access: Open.

Unified ELO 1725 ± 22, rank #227 of 1776 rated models, from 15 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
RewardBench 2 Focus93.64Accuracy (%)100
GSM-MC99.47Accuracy (%)97.8
MATH-MC Level 599.53Accuracy (%)78.7
RewardBench 2 Precise IF68.91Accuracy (%)76.5
MATH-MC Level 499.33Accuracy (%)75
JudgeBench Coding96.43Accuracy (%)73.5
RewardBench 2 Safety87.56Accuracy (%)68.6
MATH-MC Level 398.92Accuracy (%)65.4
RewardBench 2 Math88.93Accuracy (%)63.7
JudgeBench Math89.29Accuracy (%)60.8
RewardBench 2 Factuality76.68Accuracy (%)58.8
JudgeBench Knowledge77.6Accuracy (%)56.9

Interactive version: theaggregate.ai/model?slug=deepseek-v4-flash-fp8 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.