Key facts
- Mistral AI released Mistral Large 4, a 1T parameter model nicknamed 'Le Chonk'.
- The model uses a mixture of experts design, activating 49 billion parameters per query.
- Mistral plans to release the model's weights by the end of October.
- Mistral Large 4 is priced at $1.36 per million input tokens and $4.18 per million output tokens.
- Mistral AI raised €3 billion ($3.37 billion) in a Series D funding round at a valuation above €21 billion ($23.6 billion).
Mistral AI, a Paris-based artificial intelligence lab, has launched Mistral Large 4, a new AI model with one trillion parameters. The model, nicknamed 'Le Chonk' after a popular internet meme, is designed to compete with both open and closed-weight AI rivals. Mistral Large 4 utilizes a 'mixture of experts' architecture, activating 49 billion parameters per query, making it more efficient to run than models with all parameters active simultaneously.
Mistral AI Chief Scientist Guillaume Lample stated that Large 4 is "at the frontier of open models, and by far the strongest open-weight model from the US or Europe." The company plans to release the model's weights by the end of October, allowing external developers to download and test it. Mistral aims to provide 'sovereign AI' solutions, enabling companies and governments to run models without sharing data externally, a service for which Saudi Arabia's HUMAIN signed a multi-hundred-million-euro deal in August.
The company recently raised €3 billion ($3.37 billion) in a Series D funding round, achieving a valuation above €21 billion ($23.6 billion), led by Samsung. Mistral positions Large 4 as the first milestone funded by this investment.
Pricing for Mistral Large 4 is set at $1.36 per million input tokens and $4.18 per million output tokens, significantly undercutting competitors like Claude Opus 5.5 and GPT-6 Astra. Benchmarks provided by Mistral show Large 4 performing competitively against other open-weight models in areas such as coding and general task completion, though it trails some closed models in specific evaluations.
