Definition
Multimodal AI models process and reason across multiple input types — typically text and images — in a unified architecture rather than chaining separate vision and language models.
Key Points
- 2026-10-06: mistral-large-4 natively multimodal MoE — 1.6B vision encoder, 52B active params, 1M context (2026-10-07-mistral-large-4-docs)
- Visual grounding claims exceed some frontier closed models (vendor-reported)
- Open-weight multimodal at trillion-parameter scale enables on-premises document and image analysis