DeepSeek V3.2 — Benchmarks

Benchmark scores for DeepSeek V3.2 aggregated from public leaderboards, with how it ranks among open models. See hardware requirements for what you need to run it.

Overall rank: #50 of 123 open modelscomposite 51.8/100 across 9 benchmarks in 4 categories · methodology

Coding

Benchmark	Score	Open rank	All models
SWE-bench Lite	30.7%	#3 / 3	#46 / 80
Terminal-Bench	39.6%	#5 / 16	#33 / 57
SWE-bench Verified	70.0%	#5 / 16	#44 / 162
SWE-bench Bash Only	60.0%	#3 / 9	#26 / 48
SWE-bench Multilingual	59.0%	#4 / 4	#12 / 13

Knowledge

Benchmark	Score	Open rank	All models
MMLU-Pro	85.0%	#6 / 119	#33 / 259

Math

Benchmark	Score	Open rank	All models
FrontierMath	22.1%	#4 / 12	#36 / 101
ProofBench	8.0%	#14 / 21	#54 / 64

Reasoning

Benchmark	Score	Open rank	All models
ARC-AGI	57.0%	#9 / 16	#103 / 200

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