French artificial intelligence company Mistral has released a preview version of ML4, a new large-scale model that can already be tested through the Mistral Studio platform. The company plans to make the model’s weights openly available on 27 October.
ML4 was trained from scratch over two months on 4,000 graphics processors in Mistral’s European data centres. The model works in more than 160 languages, including all the official languages of the European Union, in both Latin and non-Latin alphabets.
According to Mistral, the model excels in cybersecurity, industry, finance and multimodal tasks. Co-founder and chief scientist Guillaume Lample described it as “a new generation of models”, able to help companies and governments defend themselves against cyber attacks.
The performance results are reported by the company itself and training is not yet fully complete. The launch strengthens the push for artificial intelligence developed in Europe.
