Skip to main content
Language modelActive

Mistral Large 3

Mistral AI

Released
December 2, 2025
Data date
September 8, 2026
Benchmark-Meta-Score
Unavailable
Minimum coverage not met
Speed
76.6 tokens/s
Output throughput
Price
$0.5 · $1.5
Input · Output / 1M tokens
Input
Text, Image
Output
Text

New in Mistral Large 3 is Mistral’s first mixture-of-experts checkpoint since the Mixtral line. Mistral describes 675 billion total parameters and 41 billion active parameters with a 256,000-token context window. Text and images are inputs, and the weights use Apache 2.0.

Large 3 is released as an open checkpoint. Ministral 3 14B belongs to the same multimodal Mistral 3 release alongside smaller variants.

Specifications and access

SpecificationValue and source
Weights
Open weightsSource
Access
API, LocalSource
Parameters
675BSource
Active parameters
41BSource
Context window
262,144 tokensSource
Notes
Sparse MoE with 675B total, 41B activeSource
License
Apache-2.0Source
Release
2025-12-02Source
Parameters
675 BSource
Active parameters
41 B (HF card header describes 673B language-model parameters and 39B active plus vision encoder; source descriptions differ.)Source
Architecture
sparse mixture-of-experts language model with vision encoderSource
Input
Text, ImageSource
Output
TextSource
Training hardware
3000 H200 GPUsSource
Tool calling
YesSource