benchgap
DeepSeek · model

DeepSeek V4.1 Flash (max) benchmark scores

As of 2026-10-07, DeepSeek V4.1 Flash (max) (DeepSeek) has measured scores on 9 benchmarks and estimated scores on 6 more.

BenchmarkScoreSource
Terminal-Bench 2.186.9%estimated ± 4.0 pp, high confidence
Terminal-Bench 4.026.8%measured
Terminal-Bench Science 0.19.0%measured
GPQA Diamond91.4%estimated ± 1.3 pp, high confidence
SciCode51.9%measured
CritPt14.3%measured
GDP.pdf12.8%measured
τ³-Bench Banking46.7%estimated ± 2.6 pp, medium confidence
MMMU-Pro77.0%measured
AA-AnalystAgent54.8%estimated ± 7.8 pp, medium confidence
Harvey LAB93.5%estimated ± 1.6 pp, low confidence
Humanity's Last Exam39.2%measured
AutomationBench68.9%measured
EnterpriseOps-Gym42.4%estimated ± 6.2 pp, low confidence
ITBench SRE46.9%measured