Title: DeepSeek-V3.2 %E2%80%94 deepseek-v3-2
Canonical: https://llmeu.com/lt/models/deepseek-v3-2/
Markdown mirror: https://llmeu.com/lt/models/deepseek-v3-2.md
Machine-readable: https://llmeu.com/llms.txt (index) - https://llmeu.com/llms-full.txt (full catalogue) - https://llmeu.com/llms-catalogue.json
Provider: DeepSeek
Licence: mit
Hosting: catalogue only %E2%80%94 not hosted on LLM EU yet
Sources: https://huggingface.co/deepseek-ai/DeepSeek-V3.2

# DeepSeek-V3.2

[DeepSeek](/lt/providers/deepseek/)

Paskutin%C4%AF kart%C4%85 per%C5%BEi%C5%ABr%C4%97ta: 2026-10-14

## Suverenitetas ir talpinimas

%C5%A0is modelis yra kataloge tik nuorodai. Jis dar n%C4%97ra talpinamas LLM EU, ir joks pirmosios %C5%A1alies galinis ta%C5%A1kas jo neaptarnauja. U%C5%BEklausos jam siun%C4%8Diamos tiesiogiai teik%C4%97jui, pagal to teik%C4%97jo s%C4%85lygas.

## Santrauka

DeepSeek-V3.2 yra 685 mlrd. ekspert%C5%B3 mi%C5%A1inio modelis, i%C5%A1leistas pagal MIT licencij%C4%85, su maksimalia 163 840 %C5%BEeton%C5%B3 konfig%C5%ABracija. Pagrindinis jo architekt%C5%ABrinis pakeitimas yra DeepSeek Sparse Attention, kur%C4%AF pardav%C4%97jas apib%C5%ABdina kaip suma%C5%BEinant%C4%AF ilgalaikio konteksto i%C5%A1vad%C5%B3 skai%C4%8Diavimo s%C4%85naudas. Tai yra tik teksto modelis ir jam reikalingas keli%C5%B3 mazg%C5%B3 diegimas visu tikslumu.

## Specifikacijos

- Teik%C4%97jas
  DeepSeek (CN)
- Konteksto langas
  164k
- Parametrai
  685.36B
- Modalumai
  text
- Licencija
  mit %C2%B7 [licence](https://opensource.org/license/mit)
- Atvirieji svoriai
  yes
- I%C5%A1leidimo data
  2025-12-01
- U%C5%BEduotys
  [M%C4%85stymas](/lt/tasks/reasoning/), [Programavimas](/lt/tasks/coding/), [Apibendrinimas](/lt/tasks/summarization/), [Atk%C5%ABrimas (RAG)](/lt/tasks/rag/)

## Kalbos

**Stipri:** en, zh

**Pakankama:** de, fr, es, it, pt, ru, ja, ko, ar

#### Stipri

- en
- zh

#### Pakankama

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

Nepatvirtinta: teik%C4%97jas nepagrind%C4%97 %C5%A1io teiginio. %E2%80%94 a language is listed as strong only where a source claims it; the card links the evidence where one exists.

## Kada naudoti

#### Stipryb%C4%97s

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

#### Apribojimai

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

## Da%C5%BEniausiai u%C5%BEduodami klausimai

### Is DeepSeek-V3.2 served on LLM EU?

It has no endpoint in the catalogue, so it is catalogue only %E2%80%94 not hosted on LLM EU yet, with no region and no retention to report. A card whose status is "listed" is a catalogue entry, not a running deployment.

### Under what licence is DeepSeek-V3.2 released?

DeepSeek-V3.2 is recorded under mit. The recorded licence text is https://opensource.org/license/mit. The weights are downloadable (open_weights is true). LLM EU republishes the licence id and the link and gives no legal advice; the licence text governs.

### How large is DeepSeek-V3.2, and what context window does it have?

DeepSeek-V3.2 is recorded at 685B (safetensors total 685,355,329,792) (source: https://huggingface.co/api/models/deepseek-ai/DeepSeek-V3.2), with a context window of 163840 tokens (source: https://huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/main/config.json).

### Kainodara

%C5%A0is modelis yra kataloge tik nuorodai. Jis dar n%C4%97ra talpinamas LLM EU, ir joks pirmosios %C5%A1alies galinis ta%C5%A1kas jo neaptarnauja. U%C5%BEklausos jam siun%C4%8Diamos tiesiogiai teik%C4%97jui, pagal to teik%C4%97jo s%C4%85lygas.

### Patvirtinimo b%C5%ABsena

Kai laukas neturi %C5%A1altinio, kortel%C4%97je jis rodomas kaip nepatikrintas, o ne u%C5%BEpildomas. Tu%C5%A1%C4%8Dias laukas yra informacija.

LLM EU Verified rei%C5%A1kia dokumentuota, o ne teisi%C5%A1kai sertifikuota. Tai n%C4%97ra AI akto atitikties vertinimas, ir LLM EU n%C4%97ra ES institucija.

### Versijos ir variantai

- catalogue %C2%B7 Tik kataloge %C2%B7 [weights](https://huggingface.co/deepseek-ai/DeepSeek-V3.2)

### U%C5%BEklausos pavyzdys

```
curl https://api.llmeu.com/v1/chat/completions \
  -H "Authorization: Bearer $LLMEU_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"deepseek-v3-2","messages":[{"role":"user","content":"Hello"}]}'
```

%C5%A0iam modeliui n%C4%97ra %C4%AFdiegta bandym%C5%B3 aik%C5%A1tel%C4%97: joks galinis ta%C5%A1kas jo neaptarnauja, tod%C4%97l u%C5%BEklausa netur%C4%97t%C5%B3 kur nueiti.

### Etaloniniai testai

N%C4%97ra paskelbto vertinimo. Tai spraga, o ne pasl%C4%97ptas rezultatas.

### %C5%A0altiniai

- [huggingface.co](https://huggingface.co/deepseek-ai/DeepSeek-V3.2)

### Kortel%C4%97s pakeitim%C5%B3 %C5%BEurnalas

- 2025-12-01 %E2%80%94 I%C5%A1leidimo data
- 2026-10-14 %E2%80%94 Paskutin%C4%AF kart%C4%85 per%C5%BEi%C5%ABr%C4%97ta

Paskutin%C4%AF kart%C4%85 per%C5%BEi%C5%ABr%C4%97ta: 2026-10-14