DeepSeek-V3.2

DeepSeek

Samo v katalogu Odprte uteži

Suverenost in gostovanje

Samo v katalogu

Ta model je v katalogu za referenco. Na LLM EU še ni gostovan in nobena lastna končna točka ga tukaj ne streže. Zahteve zanj gredo neposredno k ponudniku, pod lastnimi pogoji tega ponudnika.

Povzetek

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.

Specifikacije

Ponudnik
DeepSeek (CN)
Kontekstno okno
164k
Parametri
685.36B
Modalnosti
text
Licenca
mit · licence
Odprte uteži
yes
Datum izdaje
01. dec. 2025
Naloge
reasoning, coding, summarization, rag

Jeziki

Močno: en, zh

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

Jeziki

Močno

  • en
  • zh

Zadostno

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

Ni preverjeno: ponudnik te trditve ni dokumentiral. — a language is listed as strong only where a source claims it; the card links the evidence where one exists.

Kdaj ga uporabiti

Prednosti

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

Omejitve

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