Open models you can run on your own infrastructure

Pick a model. Each page covers workload fit, GPU requirements, and how to deploy it.

Mistral

Mistral

Flagship general-purpose multimodal lineup spanning Small, Medium, and Large tiers for complex reasoning and creation.

Ministral

Ministral

Compact 3B-14B multisize models balancing efficiency with strong text and vision quality.

Magistral

Magistral

Reasoning-tuned multimodal experts aimed at more deliberate analytical tasks.

Devstral

Devstral

Agentic SWE-focused family for code understanding, editing, and automation.

Voxtral

Voxtral

Speech-centric models for transcription, realtime captioning, and audio-first interfaces.

Mistral Nemo

Mistral Nemo

Multilingual open-weight model optimized for broad language coverage.

Not sure which model fits?

The private LLM study maps 29 workloads across six patterns and shows where each model family fits.

Looking at the infrastructure side?

GPU provider options, deployment architecture, and how we manage the serving layer on Kubernetes.