DeepSeek V4.1 Flash: benchmark results

Provider: DeepSeek. Released 2026-09-10. Access: Open.

Unified ELO 1966 ± 10, rank #29 of 2656 rated models, from 190 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Humanity's Last Exam (Self-Reported, With Tools)63.9Accuracy (%)100
LLM Stats (AutomationBench)54.8Score (%)100
LLM Stats (CyberGym)88.1Score (%)100
LLM Stats (NL2Repo)64Score (%)100
LLM Stats (Terminal-Bench 2.1)90.6Score (%)100
LLM Stats (ZEROBench)49Score (%)100
Nejumi 4 - jaster (0-shot) - NIILC63.01Character F1 (x100)100
PLCC - Grammar98Accuracy (%)100
SecIT Bench (Pydantic AI)82.9Accuracy (%)100
Nejumi 4 - jaster (2-shot) - JaMP84Exact match (%)98.5
Featherbench100Pass Rate (%)97.6
LLM Stats (DeepSWE 1.1)74.2Score (%)96.8

Interactive version: theaggregate.ai/model?slug=deepseek-v4-1-flash · How It Works · Data refreshed daily, snapshot 2026-09-19.