All AI models
Open sourceVerified August 11, 2026

Mistral AI

Mistral Large 3

Mistral Large 3 is a permissively licensed frontier-scale model family for teams that want strong general and multilingual capability with image understanding and the option to customize. Its operational reality is closer to a multi-GPU platform than a local model, so choose it for control and scale rather than a cheap first experiment.

License
Apache 2.0
Context
Verify selected endpoint or checkpoint
Modalities
Text · Image understanding
Parameters
675B total, 41B active

What the license means

Mistral released the base and instruction versions of Large 3 under Apache 2.0, making the weights broadly usable and adaptable under a permissive open-source license.

Open-weight, open source, and no-cost API access mean different things. Always review the linked terms and your own data, hosting, and compliance requirements before shipping.

API and operating cost

Input / 1M
Check provider
Output / 1M
Check provider

The release source confirms availability across Mistral and partner platforms but does not provide a current universal token price for this exact release.

Check current provider price

Where it fits

  • Enterprise custom models
  • Multilingual assistants
  • High-throughput multi-GPU inference

Where it does not fit

  • Local workstation deployment
  • Teams that need a published universal API rate before testing

Local deployment reality

Mistral documents optimized NVFP4 serving on Blackwell NVL72 systems and multi-GPU A100 or H100 nodes. This is a cloud or data-center open model, not a typical workstation deployment.

vLLMSGLangTensorRT-LLM

Practical setup notes

  • Use the provider's compressed checkpoint and supported runtime for the target hardware.
  • Evaluate the instruction checkpoint separately from any forthcoming reasoning variant.

Capability notes

Reasoning
Instruction release, with reasoning variants handled separately by the provider
Tools and output
Supported through Mistral and open serving ecosystem integrations

Limitations to plan around

  • Deployment needs large-scale GPU infrastructure.
  • Provider and partner pricing are not a single canonical rate.

Continue comparing

Related model decisions

See all reviewed models
Open-weight

MiniMax M3

Agentic coding

Open source

GLM-5.2

Long-horizon coding agents

Open source

Mistral Small 4

Multimodal document analysis

Editorial note: This is a dated decision page, not a benchmark leaderboard or a substitute for legal, security, or infrastructure review. Model and pricing facts can change after August 11, 2026; use the linked primary sources before committing budget or customer data.