DeepSeek · model

DeepSeek V4 Pro 0813 (max) benchmark scores

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, DeepSeek V4 Pro 0813 (max) (DeepSeek) has measured scores on 1 benchmark and estimated scores on 3 more.

Benchmark	Score	Source
τ³-Bench Banking	50.1%	estimated ± 11.4 pp, low confidence
AA-AnalystAgent	45.7%	estimated ± 3.6 pp, medium confidence
Harvey LAB	94.0%	estimated ± 1.2 pp, medium confidence
EnterpriseOps-Gym	49.6%	measured