Mistral AI
EU SovereignEuropean-sovereign open-weight models. Mistral Large 3 leads the open frontier.
Mistral AI continues to lead the open-weight market in Europe and is the clearest answer to the EU AI Act's data-residency pressure. Mistral Large 3 is currently the only open-weight model competitive with the cloud-only frontier in multimodal reasoning and instruction following. Mistral Med 3.5 hits 91.4% MMLU on the Telecom benchmark and 77.6% SWE-bench at a 128B dense parameter count — strong intelligence-per-parameter for self-hosted deployments.
Models
Mistral Large 3
FlagshipOnly open-weight model competitive with cloud frontier
Context Window
128K
Max Output
32K tokens
Input Price
Self-host or research license
Input Types
Text, Images
Output Types
Text, Code, JSON
Mistral Med 3.5
91.4% MMLU (Telecom), 77.6% SWE-bench
Context Window
128K
Max Output
16K tokens
Input Price
128B dense weights
Input Types
Text
Output Types
Text, Code
Mistral Small 3
Ultra-fast, cost-efficient default for volume
Context Window
32K
Max Output
8K tokens
Input Types
Text
Output Types
Text, Code
Codestral
Best-in-class code generation across 80+ languages
Context Window
256K
Max Output
32K tokens
Input Types
Text, Code
Output Types
Code
Pixtral Large
Multimodal Mistral for vision tasks
Context Window
128K
Max Output
8K tokens
Input Types
Text, Images
Output Types
Text
Use Cases
EU AI Act Compliance
Self-host Mistral on EU infrastructure to meet GDPR and AI Act data-residency requirements without losing frontier capability.
Air-Gapped Deployment
Open weights mean Mistral can run inside fully isolated environments — defence, finance, healthcare with strict data isolation.
Code Automation
Codestral's 256K context handles full-codebase refactors, automated PR review, and test generation.
High-Volume Text
Mistral Small handles millions of classification, extraction, and summarisation tasks at very low cost.
Why use Mistral AI on Nagent?
Nagent adds enterprise orchestration, observability, and workflow automation on top of Mistral AI's raw model capabilities.
EU data residency guaranteed — no data leaves European infrastructure
Open weights for fully air-gapped self-hosted deployments
Codestral is wired into our developer-agent skill for CI/CD pipelines
Function-calling natively supported in our tool framework
How to access Mistral AI on Nagent
Open Agent Studio
Navigate to Agent Studio in your Nagent workspace.
Select Mistral AI
Choose Mistral and pick your model. For sovereign deployments, select the self-hosted option.
Configure Data Residency
If required, enable EU-only routing in your workspace settings before deploying.
Common questions about Mistral AI
Real buyer and developer questions, answered. Click any item to expand.
Why pick Mistral if I can use GPT-5 or Claude?
Two reasons: EU AI Act compliance (data must remain on EU infrastructure) and air-gapped self-hosting (open weights). For everything else, the closed-model frontier is currently stronger on raw capability.
What does "sovereign AI" actually mean in practice?
Your model weights and inference run on infrastructure you control — not the model provider's servers. This is required for some regulated industries, most defence work, and a growing list of EU-public-sector tenders.
Is Codestral really better than GPT-5.3 Codex for coding?
On 80+ language coverage and 256K context for codebase-wide refactors, yes. For interactive single-file work, GPT-5.3 Codex still has an edge. Use Nagent's router to pick per-task automatically.
Can I mix Mistral with closed-model providers in one agent?
Yes — our tier-aware router can default to closed-model frontier for hard tasks and fall back to Mistral when data residency requires it. The agent definition stays the same; the router handles provider selection.
Ready to use Mistral AI inside your agents?
Get started in minutes — no API key management required.