Core capabilities
- 256K multimodal context for text and image work
- Function calling, structured and predicted outputs
- Agents, built-in tools, OCR and document Q&A
- Open weights plus managed API access
← Model indexMODEL PROFILE / mistral-medium-3-5
Mistral AI / open
Mistral Medium 3.5
An open-weight frontier multimodal model optimized for agentic and coding use cases.
Context256K
ModalityText · Image
AccessWeights · API
ReasoningHybrid
Where it fits.
Compelling for teams that want deployment control without giving up strong general and visual capability. Medium 3.5 is Mistral's frontier-class open multimodal model for agents and coding. The official catalog lists 128B parameters, 256K context, modified-MIT weights and a wide API feature surface.
Best-fit workflows
- Private deployment
- Coding agents
- Multimodal enterprise work
04 / OPERATIONSFrom evaluation
From evaluation
to production.
Deployment notes
Choose managed API for fast adoption or self-host when control and residency justify the operational load. Benchmark the actual runtime, quantization and tensor parallel setup; add moderation and observability as separate services.
Watch before choosing
Open weights transfer governance, inference optimization, safety filtering and capacity planning to the deployer. Open weights transfer inference optimization, safety filtering and capacity planning to the operator. Verify license terms and hardware memory needs for the exact quantization before committing to self-hosting.
05
Family & versions
- Mistral Medium 3.5Open frontier
- Mistral Small 4Efficient hybrid
- Devstral 2Software engineering
Specifications and positioning checked against provider documentation.
Open official source↗