DeepSeek V4 Pro (0813) (High): benchmark results

Provider: DeepSeek. Released 2026-08-13. Access: Open.

Unified ELO 1714 ± 1, rank #102 of 3078 rated models, from 11 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
WebDev Arena1581.64Arena Score88.1
ObviousBench98.61Answer pass³ (%)86.9
ARC-AGI-259.72Accuracy (%)73.5
Multi-turn Debate (Lechmazur)1579.7Bradley-Terry Rating72.5
ARC-AGI-187.17Accuracy (%)72
Context Arena64.09Average Score (%)70.1
AI Coding Daily (OpenCode) - Laravel Code Quality17.25Laravel Code Quality (max 20) points, LLM-judged rubric scor63.2
Computer Anthology Terminal Tasks (Terminus-2)32.6pass@1 (%)55.6
AI Coding Daily (OpenCode) - Total45.63Total points (max 60)47.4
DuelLab Overall34.2DuelLab Score45.2
AI Coding Daily (OpenCode) - React-TS Code Quality16.33React-TS Code Quality (max 20) points, LLM-judged rubric sco31.6

Interactive version: theaggregate.ai/model?slug=deepseek-v4-pro-0813-high · How It Works · Data refreshed daily, snapshot 2026-09-19.