Mistral AI
Open-weight Apache-2.0 models with an EU-hosted API — the exit ramp stays open.
70
sovereigntyMistral is the middle path between a closed API and full self-hosting: many of its models (Ministral 3B/8B/14B, Small, Devstral, Magistral Small and others) ship as open weights under Apache-2.0, so anything you build on La Plateforme's API can later move onto your own hardware — the exit ramp is built in. The API is EU-hosted (a real advantage for GDPR-sensitive workloads) and aggressively priced, from $0.04 per million tokens for Ministral 3B up to $2/$6 for Mistral Large.
OPEN SOURCEApache-2.0 (open-weight models); proprietary APISELF-HOST
LicenseApache-2.0 (open-weight models); proprietary API
PricingOpen weights free to self-host; API from $0.04/1M tokens (Ministral 3B), Large at $2/$6 per 1M
Open sourceYes
Self-hostableYes
Local-first dataNo
What it does well
- +Open Apache-2.0 weights — you can take the model home
- +EU data residency (GDPR-friendly by default)
- +Very competitive pricing across the range
- +Codestral for code, Ministral for edge/cheap workloads
Where it falls short
- −Top-end quality trails the leading US frontier models
- −Which models are open vs API-only varies — check per model
Mistral AI as an alternative to
Where Mistral AI shows up in our comparisons, and how it ranked.
Mistral AI head-to-head
Straight comparisons against the tools people weigh it against.