DeepSeek V3.1 — Benchmarks

Benchmark scores for DeepSeek V3.1 aggregated from public leaderboards, with how it ranks among open models. See hardware requirements for what you need to run it.

Overall rank: #17 of 78 open modelscomposite 55.5/100 across 5 benchmarks in 3 categories · methodology

Coding

BenchmarkScoreOpen rankAll models
WeirdML38.4%#27 / 46#115 / 161

Instruction Following

BenchmarkScoreOpen rankAll models
LMArena Text1417.3#34 / 182#118 / 391

Knowledge

BenchmarkScoreOpen rankAll models
MMLU-Pro84.8%#9 / 144#35 / 257

Reasoning

BenchmarkScoreOpen rankAll models
DTBench82.7%#18 / 67#97 / 210
LMCA24.3%#29 / 53#129 / 172
SimpleBench40.0%#14 / 25#72 / 101
Fiction.LiveBench52.8%#16 / 24#40 / 58

Scores aggregated from public benchmark sources (each linked from the benchmark pages). llmrun does not run these benchmarks.