Mistral Large 3 (675B)

Mistral AI · Pakkujad EU

Ainult kataloogis Avatud kaalud Euroopa pakkuja

Suveräänsus ja majutamine

Ainult kataloogis

See mudel on loetletud viitamiseks. Seda ei majutata veel LLM EU-s ega teeninda seda siin ükski esimese osapoole lõpp-punkt. Päringud suunatakse otse pakkujale, pakkuja enda tingimustel.

Kokkuvõte

Mistral Large 3 is the largest published Mistral model: a granular mixture-of-experts model with 675B total parameters and 41B active parameters, released under Apache-2.0. It is multimodal and supports a 256k-token context window. The vendor states it was trained from scratch on 3000 H200 GPUs.

Tehnilised andmed

Pakkuja
Mistral AI (FR)
Konteksti aken
262k
Parameetrid
675B
Modaalsused
text, vision
Litsents
apache-2.0 · licence
Avatud kaalud
yes
Avaldamiskuupäev
28. nov 2025
Ülesanded
reasoning, coding, multilingual-support, vision, summarization

Keeled

Tugev: en, fr, es, de, it, pt

Piisav: nl, zh, ja, ko, ar

Keeled

Tugev

  • en
  • fr
  • es
  • de
  • it
  • pt

Piisav

  • nl
  • zh
  • ja
  • ko
  • ar

Verifitseerimata: pakkuja ei ole seda väidet dokumenteerinud. — a language is listed as strong only where a source claims it; the card links the evidence where one exists.

Millal kasutada

Tugevused

  • frontier-scale open weights under a permissive licence
  • 41B active parameters keep decoding cheaper than the total size suggests
  • 256k-token context window
  • vendor-published NVFP4 checkpoint for single-node deployment

Piirangud

  • 675B parameters require multi-GPU nodes even quantised to NVFP4
  • no independent evaluation has been run by LLM EU
  • vendor benchmark comparisons are self-reported