Identity
- Status
- active
- Developer ID
mistral-large-2512- Released
- Dec 2, 2025
- Version
- Mistral Large 3
- License
- Apache-2.0
Mistral Large 3 is an open-weight multimodal mixture-of-experts model with 675B total and 41B active parameters.
| Provider | Input / 1M | Output / 1M | Cache read / 1M | Cache write / 1M | Observed | Evidence |
|---|---|---|---|---|---|---|
Mistral AIAvailable | $0.50 | $1.50 | $0.050 | — | Aug 29, 2026 | Provider-reported |
OpenrouterAvailable | $0.50 | $1.50 | $0.050 | — | Sep 2, 2026 | Provider-reported |
| Mistral Large 3Mistral AI | LMArena Text Arenatext-2026-09-01-011508720696 | 1,427.62 | — | 65,336 | mistral-large-3; 95% CI [1424.57601414, 1430.66670291]; votes 65336; rank 85unknown | Original sourcewinner eligible | Third-party benchmark |
| Mistral Large 3Mistral AI | LMArena Vision Arenavision-2026-08-27-011508720696 | 1,226.84 | — | 4,751 | mistral-large-3; 95% CI [1216.90651393, 1236.78308197]; votes 4751; rank 66unknown | Original sourcewinner eligible | Third-party benchmark |
| Mistral Large 3Mistral AI | ToneBench2026-08-14-10-task | 70.61 | 2,014 | 50 | Mistral Large 3average per case | Original sourcewinner eligible | Third-party benchmark |
At a Glance
mistral-large-2512This page represents one immutable developer model ID. Serving endpoints, pricing, regions, quantizations, and service tiers attach separately and never create duplicate model records.
Primary Evidence
Comparable Peers
| A | Pair | B | Context |
|---|---|---|---|
Mistral Large 3Mistral AI | vs | Hy4 previewTencent | cross-developer peerstext |
Mistral Large 3Mistral AI | vs | GLM-5.3-FlashZ.ai | cross-developer peersimage, text |
Mistral Large 3Mistral AI | vs | Nova 2 LiteAmazon | cross-developer peersimage, text |
Mistral Large 3Mistral AI | vs | Mistral-Medium-3.5-128BMistral AI | family variantsimage, text |
Mistral Large 3Mistral AI | vs | Mistral-Small-4-119B-2603Mistral AI | family variantsimage, text |
Mistral Large 3Mistral AI | vs | Command A VisionCohere | cross-developer peersimage, text |
Mistral Large 3Mistral AI | vs | DeepSeek-V4-Flash-Vision-ExpDeepSeek | cross-developer peersimage, text |
Mistral Large 3Mistral AI | vs | cross-developer peersimage, text | |
Mistral Large 3Mistral AI | vs | cross-developer peersimage, text | |
Mistral Large 3Mistral AI | vs | cross-developer peersimage, text | |
Mistral Large 3Mistral AI | vs | Gemini 3.1 ProGoogle DeepMind | cross-developer peerstext |
Mistral Large 3Mistral AI | vs | stable-diffusion-3.5-largeStability AI | cross-developer peersimage |
Questions
Mistral Large 3 is an open-weight multimodal mixture-of-experts model with 675B total and 41B active parameters. It is developed by Mistral AI and its current sourced lifecycle status is active.
Mistral Large 3's current record lists Text, Image, and Document as input and Text as output.
The current record lists a 262,144-token context window and does not report a maximum output length.
Mistral Large 3 is marked as API-available. The directly sourced provider records currently include Mistral AI and Openrouter.
The current record lists 675,000,000,000 total parameters and 41,000,000,000 active parameters. Its architecture is mixture_of_experts.
Mistral Large 3's recorded capabilities are agents, chat, generation, structured outputs, tools, and vision. Its supported tasks are document-question-answering, image, and text.
The lowest directly sourced prices currently attached to Mistral Large 3 are $0.50 per million input tokens through Mistral AI and $1.50 per million output tokens through Mistral AI. Prices are provider-specific and should be checked against each cited observation date.
Mistral Large 3 is an open-weight model. Self-hosting is supported, and the recorded license is Apache-2.0.
No predecessor is recorded. No successor is recorded. Recorded aliases are mistral-large-2512 and mistral-large-3.
The current compatible peer set includes Hy4 preview, GLM-5.3-Flash, Nova 2 Lite, Mistral-Medium-3.5-128B, and Mistral-Small-4-119B-2603. Compatibility requires shared sourced modalities and tasks; it does not imply that one model is better. Browse model comparisons →