Mistral Large

The Mistral Large family represents Mistral AI's flagship line of large language models, evolving through successive releases that progressively expand capability and scale. The lineage continued with Mistral Large the listed price, officially announced on 2024-07-24, which was built around a 123 billion parameter architecture and engineered for efficient single-node inference. Training emphasized a substantial proportion of code data, drawing on the team's experience with earlier code-focused models like Codestral 22B and Codestral Mamba, which helped the model achieve strong results across reasoning and programming benchmarks. On MMLU, the pretrained version of Mistral Large the listed price reached 84.0% accuracy, establishing a new point on the performance-versus-cost Pareto frontier among openly described models at the time of release.

Mistral Large the listed price was designed with multilingual breadth in mind, supporting dozens of natural languages including French, German, Spanish, Italian, Portuguese, Arabic, Hindi, Russian, Chinese, Japanese, and Korean, alongside 80 or more programming languages such as Python, Java, C, C++, JavaScript, and Bash. The model was released under the Mistral Research License (MRL-0.1), permitting use and modification for research and non-commercial purposes, with commercial self-deployment requiring a separate commercial license. The family has continued to advance beyond that release, with an independent review documenting a Mistral Large 3 iteration tagged 2512, suggesting ongoing investment in scaling capability. This progression makes the Mistral Large line a practical choice for teams needing strong reasoning, broad language coverage, and mature code generation within a single model family.