Abstract:
Mistral launched a new AI model on Tuesday that the company said ranks among the top open source models in the world. According to the French company, Mistral Large 4 (referred to as ML4, codenamed le Chonk) is a large model with 1 trillion parameters, which performs particularly well in network security, code writing, manufacturing, finance, and multi-modal related tasks.
Mistral has long been regarded as the only company in Europe that can benchmark OpenAI and Anthropic. The company completed 3 billion euros (equivalent to 3.4 billion U.S. dollars) in Series D financing in September, with a valuation of 21 billion euros. This round of financing solidifies its position as one of the best-funded and highest-valued AI startups in Europe.
Mistral stated that Mistral Large 4 is now available in preview for developers, cybersecurity leaders, and government agencies, and will be officially released to a wider audience later this month.
The ML4 model relies on 4,000 NVIDIA Grace Blackwell graphics cards and took two months to complete training in Mistral's own data center in Europe.
Mistral said in a statement that after the core parameters are officially released, the ML4 comprehensive benchmark performance will rank among the world's top open weight models.
However, this new model still lags behind the industry's cutting-edge levels in areas such as code writing.
Guillaume Lample, co-founder and chief scientist of Mistral, said: "After completing the D round of financing, as the scale of training computing power continues to increase, the model performance will be further enhanced."
He added: "In addition, the model's cyber defense capabilities can help enterprises and governments defend against malicious attackers who will bypass the security guardrails of the closed source model and use it to carry out cyber attacks."
Comments