AI Tools.

Search
by mistralai

Mistral-Small-3.1-24B-Instruct-2503

Mistral Small 3.1 24B is Mistral AI's March 2025 update to their small-tier model family, updating to mistral-common v2 and adding multimodal vision capabilities to the 24B instruction model. It supports vision inputs alongside text and covers 25+ languages. The model uses a custom mistral3 architecture serialised for vLLM serving.

Summary text generated by an automated pipeline from the model card · Not individually reviewed or run by us · How this page is made

Share: X Reddit HN LinkedIn

From the model card

Fields below are copied from the tags and counters on the HuggingFace repository mistralai/Mistral-Small-3.1-24B-Instruct-2503 at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.

Publisher (HF namespace)
mistralai
Library
vLLM
Weight formats
safetensors
License tag
apache-2.0 — read the license file in the repo before relying on it
Lineage
Language tags
English (en), French (fr), German (de), Spanish (es), Portuguese (pt), Italian (it), Japanese (ja), Korean (ko), Russian (ru), Chinese (zh), Arabic (ar), Persian (fa), Indonesian (id), Malay (ms), Nepali (ne), Polish (pl), Romanian (ro), Serbian (sr), Swedish (sv), Turkish (tr), Ukrainian (uk), Vietnamese (vi), Hindi (hi), Bangla (bn)
Downloads (HF counter at last fetch)
336,723
Likes (HF counter at last fetch)
1,363
Model card
https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503

Use cases

  • Multilingual instruction following across 25+ languages including Asian scripts
  • Vision-language tasks such as image captioning and visual QA
  • Cost-efficient 24B-scale deployment where GPT-4o-class quality is not required
  • Enterprise multilingual chat and document understanding
  • Building applications that need both text and vision capabilities in a single model

Pros

  • 24B scale with vision at a fraction of frontier model serving cost
  • 25+ language coverage including Arabic, Hindi, Vietnamese, and CJK
  • 1361 likes; one of Mistral's most widely adopted releases
  • vLLM compatible via mistral-common tooling

Cons

  • mistral-common v2 serialisation requires specific Mistral library versions
  • No pipeline_tag on HuggingFace; inference setup is less obvious than standard models
  • Vision capabilities are new to this 3.1 update; less battle-tested than text modality
  • 24B requires multi-GPU or quantisation for consumer deployment

Tags

vllmsafetensorsmistral3mistral-commonenfrdeesptitjakoruzharfaidmsnepl