Frontier Log

Benchmark · No longer reported

MMLU-Pro

020406080100Apr 202687.5%DeepSeek-V4-Pro020406080100Apr 202687.5%DeepSeek-V4-Pro

Results

ModelScoreSourceHuman-verified
DeepSeek-V4-Pro24 Apr 2026 · DeepSeek87.5%DeepSeek V4 Preview ReleaseMax reasoning effortSource imageYes

Benchmark

huggingface.coHigher is better

TIGER-Lab, University of WaterlooIntroduced 3 Jun 2024

All benchmarks