Pixtral 12B

Pixtral 12B is a 12-billion parameter multimodal model from Mistral, notable as the lab's first architecture designed to take both text and images as input. It is distributed under the Apache 2.0 license and positioned as an open-weight release aimed at researchers, hobbyists, and developers building applied vision-and-language workflows. The combination of a relatively compact parameter count with native image handling makes it a practical choice for teams that want multimodal capability without committing to much larger frontier-scale systems.

A defining feature is the 128,000-token context window, which lets a single prompt carry long documents, multiple images, or richly interleaved visual and textual material. Images are consumed at their native resolution without a forced resizing or preprocessing step, simplifying pipelines for chart interpretation, document analysis, and graph understanding. A formally titled "Pixtral 12B" research paper exists alongside the release, signaling that the model is accompanied by a documented technical contribution rather than being a purely productized drop. This mix of open licensing, generous context, and flexible image handling makes it well suited to multimodal prototypes and mixed-modality assistants.