Mistral Large 3
Mistral Large 3 is a multimodal mixture-of-experts model from December 2025. It has about 675B total parameters and 41B active parameters. The API is deprecated, but the official instruction weights remain downloadable under Apache 2.0.
Deprecated API; check access details
Separate precision variants from new releases
The model card offers FP8, BF16 and NVFP4 versions. It describes a 673B language component with 39B active parameters and a 2.5B vision encoder. The rounded 675B and 41B headline counts include this multimodal structure. Its instruction interface supports tools and JSON.Mistral Large 3 official weight card
Separate the deprecated API from weights
The model page records API deprecation on 31 October 2026. The official weight repository remains available. API access is subject to Mistral's retirement schedule; a published model identifier does not guarantee continued service after retirement.Mistral Large 3 API lifecycle pageMistral Large 3 official weight card
Evaluate the complete task
Practical guidance. Large 3 remains relevant to a comparison that requires Apache 2.0 weights. Measure the precision variant on the actual server and document workload. An older model with suitable access terms can be useful, but its previous API name should not be treated as a promise of continuing hosted support.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Deprecated API; check access details |
| Access | API, Downloadable |
| Model identifiers | mistral-large-2512Mistral-Large-3-675B-Instruct-2512 |
| Context window | 256,000 tokens Advertised 256k context. Local capacity depends on the serving configuration. Mistral Large 3 official weight card |
| API status | Deprecated on 31 October 2026Mistral Large 3 API lifecycle page |
| Weight license | Apache 2.0Mistral Large 3 official weight card |
| Weight repository | mistralai/Mistral-Large-3-675B-Instruct-2512Mistral Large 3 official weight card |
Related models
- Mistral Large 4
Mistral AI · Mistral Large
A sparse multimodal model with a one-million-token context and 52B active parameters, including embeddings and the output layer.
Launch date not confirmedAPI · Downloadable - Mistral Small 4
Mistral AI · Mistral Small
An Apache-licensed multimodal model that combines instruction following, reasoning and coding with selectable reasoning modes.
Launch date not confirmedAPI · Downloadable