Mistral Large 3
Language modelPrevious generationOpen weightsCard summarizedA state-of-the-art, open-weight, general-purpose multimodal model
● Re-checked 12 Oct 2026 · 40 facts · 0 corrections · 1 unconfirmedSource: Mistral Large 3 675B Instruct 2512 (Hugging Face model card)
Use it via API
Where to call it
- Mistral API (Mistral AI Studio)
mistral-large-2512 - Self-host (weights)
mistralai/Mistral-Large-3-675B-Instruct-2512 - Amazon Bedrock
mistral.mistral-large-3-675b-instruct - Microsoft Foundry / Azure
Mistral Large 3
Good for
- Long document understanding
- Daily-driver AI assistants
- Agentic and tool-use workloads with native function calling and JSON output
Overview
Large 3
A state-of-the-art, open-weight, general-purpose multimodal model
- Cost
- $0.50 in
- Context
- 256K
- Input / output
- $0.50 / $1.50
- Max output
- —
- Knows up to
- —
- Released
- 2 Dec 2025
Mistral Large 3 is a general-purpose multimodal granular Mixture-of-Experts model with 41B active and 675B total parameters, trained from scratch on 3,000 NVIDIA H200 GPUs and released with base and instruct weights under Apache 2.0. It targets production assistants, retrieval-augmented systems and enterprise workflows, with native function calling, JSON output and a 256K context. Mistral lists it as behind dedicated reasoning and vision-first models in those areas. Model card intro; Key Features; launch post
mistral-large-2512- Input / output per 1M
- $0.50 / $1.50
- Context window
- 256K
- Max output
- —
- Knows up to
- —
- Reasoning
- None
- Weights
- Open · Apache 2.0
About this card. A model card is the maker's own account of a model: what it's for, how it was trained and tested, and where it falls short. This page condenses Mistral AI's Mistral Large 3 675B Instruct 2512 (Hugging Face model card). Every claim carries its source page.