Mistral AI announced this morning the general availability of its Large 3 model through the la Plateforme API. The model has 123 billion parameters and stands out for its special attention to multilingual support, with notable performance in French, German, Spanish and Italian.
Pricing is aggressive: 4 dollars per million input tokens and 12 for output, positioning it significantly below GPT-5 and Claude Opus 4.7. Benchmarks place the model at around 86.3% on MMLU, a respectable level without reaching the peaks of American competitors.
Mistral's strategy remains clear: offer a sovereign European alternative with excellent price-performance, targeting continental businesses concerned about their tech dependencies. The model will also be available in a self-hosted version for enterprise customers.