DeepSeek-V3.2

DeepSeek

Catalogue only Open weights

Sovereignty and hosting

Catalogue only

This model is catalogued for reference. It is not hosted on LLM EU yet, and no first-party endpoint serves it here. Requests to it go to the provider directly, under that provider's own terms.

Summary

DeepSeek-V3.2 is a 685B mixture-of-experts model released under the MIT licence, with a configuration maximum of 163,840 tokens. Its main architectural change is DeepSeek Sparse Attention, which the vendor describes as reducing the computational cost of long-context inference. It is a text-only model and needs a multi-node deployment at full precision.

Specifications

Provider
DeepSeek (CN)
Context window
164k
Parameters
685.36B
Modalities
text
License
mit · licence
Open weights
yes
Release date
01 Dec 2025
Tasks
reasoning, coding, summarization, rag

Languages

Strong: en, zh

Adequate: de, fr, es, it, pt, ru, ja, ko, ar

Languages

Strong

  • en
  • zh

Adequate

  • de
  • fr
  • es
  • it
  • pt
  • ru
  • ja
  • ko
  • ar

Not verified: the provider has not documented this claim. — a language is listed as strong only where a source claims it; the card links the evidence where one exists.

When to use it

Strengths

  • MIT licence
  • sparse attention aimed at lower long-context inference cost
  • 163,840-token configuration maximum

Limits

  • 685B parameters: full-precision serving needs several nodes
  • vendor efficiency claims are not independently reproduced here
  • no evaluation has been run by LLM EU