Model
Inference & Serving
Pixtral 12B
1Mistral's 2first 3open 4multimodal 5vision 6model
the six Ws · specification
W1
Who
Mistral AI released Pixtral 12B, its first vision-language model.
W2
What
A 12-billion-parameter multimodal model that interleaves images and text for natural-image reasoning and chat.
W3
Where
Distributed on Hugging Face and via Mistral's own model hub.
W4
When
Pixtral 12B was released in September 2024.
W5
Why
It extended Mistral's open-weight lineup into multimodal understanding, competing with Qwen2-VL and LLaVA.
W6
With
Runs via the Mistral inference stack or vLLM/transformers, using a new vision encoder trained from scratch.
W7
Watch
Open-weight release by Mistral AI under Apache 2.0 license.