DeepSeek V4.1 Flash (High): benchmark results

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

Unified ELO 1716 ± 1, rank #96 of 3078 rated models, from 23 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AI Coding Daily (OpenCode) - React-TS Code Quality18.67React-TS Code Quality (max 20) points, LLM-judged rubric sco100
Vals AI SkillsBench69.8Accuracy (%)100
Chess Bench LLM1294Lichess Rating84
Vals AI CyberBench78.82Accuracy (%)84
Vals AI MedScribe85.5Accuracy (%)80.6
AI Coding Daily (OpenCode) - Total48.52Total points (max 60)78.9
LLM2014 Logic 2026-0952.09Median Score68.8
Vals AI SAGE47.88Accuracy (%)68.8
Vals AI LegalBench83.28Accuracy (%)64.6
Vals AI Public Benefits Bench64.28Accuracy (%)62.9
Vals AI ProofBench54Accuracy (%)54.3
Bug Hunt Bench - VS Code Extension9Planted Bugs Fixed (out of 45)52.2

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