Mistral Large 3 (v25.12, Base/Instruct)
Mistral Large 3 (v25.12, Base/Instruct) is an AI model published by MistralAI, released on 2025-12-02, for 多模态大模型, with 6750.0B parameters, and 256K tokens context length, under the Apache 2.0 license.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
| Modality | Input | Output |
|---|---|---|
| Text | $0.50 | $1.50 |
Mistral Large 3 currently shows benchmark results led by Pinch Bench (28 / 37, score 72.20), Claw Bench (21 / 27, score 78.60). This page also consolidates core specs, context limits, and API pricing so you can evaluate the model from benchmark results and deployment constraints together.
Mistral Large 3 为 Mistral AI 发布的开放权重多模态通用模型，采用细粒度稀疏 Mixture-of-Experts（MoE）架构，41B 激活参数、675B 总参数，上下文 256K。官方同时开源 Base 与 Instruct 权重；Reasoning 变体暂未公开发布。
本条目合并了 Base 与 Instruct（两者同代 v25.12，价格一致）；Reasoning 版本尚未上线，待官方发布后再新增。
官方面向长文档理解、工具/函数调用、Agent 工作流、多语种分析与多模态推理等企业场景，并在多家平台上线。
可在 Mistral AI Studio、AWS Bedrock、Hugging Face、OpenRouter、Fireworks、Together 等渠道使用；定价（按本模型专页）：输入 $0.50 / 1M tokens，输出 $1.50 / 1M tokens；未给出缓存或超阈值附加计费。
资料：官方新闻《Introducing Mistral 3》与 Large 3 文档、HF 集合页。
欢迎关注 DataLearner 官方微信，获得最新 AI 技术推送
