Mistral AI: native-resolution Pixtral vision and multimodal evaluation resource

What changed

Trained a variable-resolution/aspect-ratio vision encoder with interleaved image/text decoding and released a multiturn multimodal benchmark plus standardized prompt/parsing evaluation tools.

Scientific scope

Trained a variable-resolution/aspect-ratio vision encoder with interleaved image/text decoding and released a multiturn multimodal benchmark plus standardized prompt/parsing evaluation tools. ViT, RoPE, multimodal decoding and MT-Bench have predecessors. Native-resolution implementation and the evaluated resource are acquired, not invention of all variable-resolution vision.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. ViT, RoPE, multimodal decoding and MT-Bench have predecessors. Native-resolution implementation and the evaluated resource are acquired, not invention of all variable-resolution vision.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Pixtral 12B
• Mistral AI: native-resolution Pixtral vision and multimodal evaluation resource — official institutional account

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction