Mistral Large 3

Claims and evidence

• Released 2 December 2025 ● Primary source[1] [4]December 2 (2025) [2]page header date

• Status Available ● Primary source[2]GA badge [5]Generalist models [3]

• Successor of Mistral Large 24.11 EditorialEditorial link along the Mistral Large line: the announcement presents Mistral 3 as the next generation and Large 3 as Mistral's most capable model to date, without naming a predecessor; Mistral Large 24.11 was the previous Large release. The models overview lists Large 3 as the alternative for the retired Large 1.0 and 2.0 endpoints, but Medium 3.5 for Large 2.1 (24.11). ● Primary source[1]

• Change · Architecture Uses a sparse mixture-of-experts architecture with 41B of 675B parameters active; Mistral calls it its first mixture-of-experts model since Mixtral.Compared with Mistral Large 24.11 ● Primary source[1]
• Change · Context length Has a context window of 256k tokens.Compared with Mistral Large 24.11 ● Primary source[2]Context

• Input text, image ● Primary source[2]Modalities [3]Key Features
• Output text ● Primary source[2]Modalities
• Feature Function calling ● Primary source[2]Features: Function Calling
• Open weights YesBase and instruct weights under Apache 2.0. ● Primary source[3] [6] [1]
• Context window 256k tokens ● Primary source[2]Context [3]Key Features: Large Context Window
• Parameters 675B total, 41B activeSparse mixture-of-experts. The Hugging Face card splits this into a 673B-parameter language model with 39B active and a 2.5B vision encoder. ● Primary source[1] [2]description [3]Key Features
• Access API, Open-weights download, cloud partner ● Primary source[1]Available Today [2]Features: API endpoints; Weights

Lineage

Variants

No variants recorded in this record.

Related AI Radar coverage

AI Radar coverage starts in June 2026; no coverage linked yet.

All model releases from Mistral AI on AI Radar →