Ministral 3

25.12 · Family: Mistral

December 2025 family of open Mistral AI models (Apache 2.0) in 3B, 8B and 14B sizes — multimodal (text + image), 256K context, built for edge and local deployment.

✓ Active✓ Public access⚖ Open sourceLLMMultimodalReasoning model📁 Mistral

Context window

256K

tokens

Parameters

3B / 8B / 14B

parameters

Release date

2 December 2025

🏢Mistral AIProducer

Access:APIDownloadHostedDeployment:💻 Local☁ Cloud📱 On-device

Overview

Ministral 3 is a family of open large language models from the French company Mistral AI, unveiled on 2 December 2025 (version 25.12) as part of the "Mistral 3" generation launch. The models are released under the Apache 2.0 license.

The family spans three sizes — 3B, 8B and 14B parameters — each available in base, instruct and reasoning variants. All models are natively multimodal (text and image input) and multilingual, with a context window of up to 256K tokens.

Ministral 3 is designed for edge and local deployment, emphasising strong performance-to-cost ratios across diverse hardware. The models support function calling, structured outputs, document question answering and batch processing.

The largest variant, Ministral 3 14B, is reported by the vendor to match the quality of the larger Mistral Small 3.2 24B model, while its reasoning variant scores 85% on the AIME 2025 benchmark. La Plateforme API prices are USD 0.10 (3B), 0.15 (8B) and 0.20 (14B) per million tokens (input and output).

Classification

LLMMultimodalReasoning model

Family: Mistral

Access & deployment

APIDownloadHosted

LocalCloudOn-device

Weights: Open source

Key parameters

📏 Context: 256K

🧩 Parameters: 3B / 8B / 14B

✓ Tools · ✓ Fine-tuning

📥 Input: text, image, documents

Technical specification

Context window

256K

tokens

Parameters

3B / 8B / 14B

parameters

License

Apache 2.0

Features:✓ Tool use✓ Fine-tuning

Modalities

⬇ Input

textimagedocuments

⬆ Output

textcodestructured_data

Capabilities and applications

Native model capabilities

Function Calling

Category: planning

Tool use

The model's ability to call external functions, APIs and tools during a conversation: calculator, search engine, code editor, database. The model decides when and how to use a tool and interprets its result.

Category: planning

Structured output

Producing data in structured formats such as JSON.

Category: structured_generation

Image understanding

Analysing and interpreting the content of images.

Category: vision

Multimodal understanding

Category: multimodal

Multilingual

Competence in many natural languages (from a few to over a hundred): understanding, generation, translation, and code-switching within a single conversation. Frontier models support a wide range of languages with comparable quality.

Category: language

Reasoning

The model's ability to reason logically and solve complex problems.

Category: reasoning

Long context

Support for large context windows — tens to hundreds of thousands (or millions) of input tokens. Enables analysis of entire codebases, long documents, and many parallel conversations without losing earlier information. GPT-5.1 supports 400,000 tokens.

Category: language

Instruction following

Precisely following instructions contained in the prompt: response format, length, style, constraints (e.g. 'reply in six words'). GPT-5.1 significantly improved this capability compared to GPT-5.

Category: language