Ministral-3-3B-Instruct-2512 published in ONNX format for browser inference
Mistral published the ONNX export of its 3.3B vision-language instruct model, optimized for browser and edge inference via Transformers.js. The image-text-to-text model supports 12 languages and is based on the Mistral 3 architecture.
PUBLISHED2025-11-24
OBSERVED2026-08-11
AGE9mo
SOURCES1
- ONNX format for Transformers.js inference in browser/edge environments
- 3.3B parameter vision-language model (image-text-to-text pipeline)
- Mistral 3 architecture, fine-tuned from Ministral-3-3B-Instruct-2512
- 12 languages: en, fr, es, de, it, pt, nl, zh, ja, ko, ar
- Referenced arxiv paper: 2601.08584
COMMUNITY
No curated reactions recorded for this event. Facts and takes are kept in separate layers — community context is added by hand, never blended into the record above.