DOOMBENCH.COM
Competitive race

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.

0 comments · 0 votesOpen discussion

Public discussion is readable by everyone. Sign in to comment, reply, or vote.

No comments yet. Start the discussion.

Send detailed or private feedback →
CURRENT ASSESSMENT · REVISION 1
TOWARD DOOM48confidence 92/100

Why it moved the index

The exact dated release combined immediate API deployment with openly downloadable weights, code, and permissive licensing. Sparse attention reduces the cost of long-context use while benchmark results retain strong reasoning and agentic tool-use performance. That lowers deployment barriers and competitive cost pressure for a capable model, although the release does not itself establish a loss-of-control incident or real-world autonomous harm.

Assessment history

  1. R1
    Toward 48 · confidence 92

    Adds an exact, previously absent September 2025 model release with verified API access and open weights.

    14 Aug 2026

Share this page

DoomBench social sharing card for DeepSeek releases open V3.2-Exp with sparse long-context attention.

  1. DoomBench assesses “DeepSeek releases open V3.2-Exp with sparse long-context attention” as evidence moving toward doom, with magnitude 48 and confidence 92 out of 100 in the competitive race category.

  2. The DoomBench assessment of “DeepSeek releases open V3.2-Exp with sparse long-context attention” is based on reporting from DeepSeek and records the editorial rationale, source quality, attribution, and revision history.

  3. DoomBench summarizes “DeepSeek releases open V3.2-Exp with sparse long-context attention” as follows: DeepSeek upgraded its API models to DeepSeek-V3.2-Exp and released MIT-licensed weights and code for the 685-billion-parameter...

    https://www.doombench.com/news/deepseek-releases-open-v3-2-exp-with-sparse-long-context-attention-2025-09-29