DeepSeek V3.2-Exp

llm

An experimental update to V3.1-Terminus introducing DeepSeek Sparse Attention for more efficient long-context training and inference at roughly the same output quality.

Metadata

orgDeepSeek
released2025-09-29
params671B-A37B
licensemit
kindllm
hf iddeepseek-ai/DeepSeek-V3.2-Exp

Benchmarks

MMLU-Pro85source ↗
GPQA79.9source ↗
AIME 202589.3source ↗
SWE-bench67.8source ↗

Lineage

↑ finetuneDeepSeek V3.1
↓ finetuneDeepSeek V3.2
view in full graph →