HF Model & Benchmark Matrix

24,078 variants · 14,984 scores · 120 benchmarks
Home/Orgs/deepseek-ai/deepseek-ai/DeepSeek-V4-Pro

deepseek-ai/DeepSeek-V4-Pro

family deepseek · 4 variants · unknown

provider
deepseek-ai
released
—
huggingface
https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro
best scores
Terminal-Bench: 2.0AGIEval: 80.1MMLU: 87.8MMLU-Pro: 65.5C-Eval: 90.4CMMLU: 88.9SimpleQA: 76.4TriviaQA: 83.3BBH: 87.6DROP: 88.2HellaSwag: 86.4WinoGrande: 78.9BigCodeBench: 63.9HumanEval: 62.8GSM8K: 91.1MATH: 60.5MGSM: 81.3LongBench: 40.2GPQA Diamond: 91.3Humanity's Last Exam: 53.1LiveCodeBench: 88.8Terminal-Bench 2.0: 65.4BrowseComp: 83.7

Best scores per benchmark

benchmarkcategorybest score
GPQA Diamondreasoning91.3
GSM8Kmath91.1
C-Evalknowledge90.4
CMMLUknowledge88.9
LiveCodeBenchcoding88.8
DROPreasoning88.2
MMLUknowledge87.8
BBHreasoning87.6
HellaSwagknowledge86.4
BrowseCompagentic83.7
TriviaQAknowledge83.3
MGSMmath81.3
AGIEvalreasoning80.1
WinoGrandeknowledge78.9
SimpleQAknowledge76.4
MMLU-Proknowledge65.5
Terminal-Bench 2.0agentic-coding65.4
BigCodeBenchcoding63.9
HumanEvalcoding62.8
MATHmath60.5
Humanity's Last Examknowledge53.1
LongBenchlong-context40.2
Terminal-Benchagentic-coding2.0

Variants (4)

variantkindquantbitsvRAMdlorgyear
DeepSeek-V4-Profp8fp88.01,648,028deepseek-ai2026
DeepSeek-V4-Pro-NVFP4nvfp4nvfp44.0100,545nvidia2026
DeepSeek-V4-Pro-Basefp8fp88.046,096deepseek-ai2026
DeepSeek-V4-Profp8fp88.0172unsloth2026
Hugging Face model & benchmark data scraped & matrix-ified. 24,078 variants · 14,984 scores · 120 benchmarks.