DeepSeek-V3.2-Exp
A 685-billion-parameter experimental successor to DeepSeek-V3.1-Terminus using DeepSeek Sparse Attention for more efficient long-context training and inference, released through DeepSeek's API with MIT-licensed weights and code.
0 comments · 0 votes
Sign in to join the discussion →
No comments yet. Start the discussion.
Why this model scores 82.4
Primary release materials show strong reasoning and agentic tool-use performance broadly comparable with V3.1-Terminus, plus an architectural efficiency improvement for long contexts. Immediate API availability, openly downloadable 685B weights, implementation code, and an MIT license justify maximal deployment exposure. The autonomy, misuse, and control-difficulty assessments remain below later DeepSeek-V3.2 variants because this experimental release did not document stronger independent autonomy or a concrete control failure.
News tied to DeepSeek-V3.2-Exp
The model score of 82.4 rates this model's risk profile. The overall Doom Index of 61.3 measures the complete temporally weighted evidence record. These values answer different questions.
Each article's current Doom Index contribution is divided equally among the exact models named on that article. This prevents multi-model evidence from being claimed in full on several model pages. Model risk scores use a bounded temporal offset around their technical profile, but never feed back into the overall index.
DeepSeek releases open V3.2-Exp with sparse long-context attention
DeepSeek upgraded its API models to DeepSeek-V3.2-Exp and released MIT-licensed weights and code for the 685-billion-parameter experimental model, adding sparse attention intended to make long-context inference more efficient while retaining V3.1-Terminus performance.
- Full item contribution
- +0.24
- DeepSeek-V3.2-Exp equal share
- +0.24
Model score history
- R1Doom Score 82.4
Adds the exact experimental 3.2 release, which is distinct from the later final DeepSeek-V3.2 and V3.2-Speciale records.
14 Aug 2026