Mistral unveils ML4, its most advanced AI model yet
What's the story
Mistral, a leading French AI start-up, has launched its new flagship AI model, Mistral Large 4 (ML4). The company claims it is one of the most advanced open models globally. The launch comes as Western companies are racing to catch up with Chinese developers in the field of open-weight and open-source artificial intelligence (AI).
Model details
ML4 excels at cybersecurity, coding, and finance tasks
Mistral Large 4, or le Chonk, is a model with a whopping one trillion parameters.
The company claims that it excels at tasks related to cybersecurity, coding, manufacturing, finance, and multimodal applications.
Mistral has been Europe's leading competitor to OpenAI and Anthropic for years now.
In September, the company raised €3 billion ($3.4 billion) in a Series D funding round at a valuation of €21 billion.
Training process
ML4 strongest open-weight model developed outside China
The training of ML4 took two months on 4,000 NVIDIA Grace Blackwell GPUs in Mistral's own data centers in Europe.
Once its core parameters are released, ML4 will be among the best open-weight models globally based on aggregate benchmark performance.
Mistral also claims that ML4 is the strongest open-weight model developed outside China by a "substantial margin."
Enhancement
Mistral to scale up training capacity after Series D fundraise
Despite its impressive capabilities, ML4 still has some areas to improve, such as coding.
However, Mistral's co-founder and chief scientist Guillaume Lample said these capabilities will improve as they scale up their training capacity after their Series D fundraise.
He also added that the cyber defense capabilities of ML4 will help enterprises and governments protect themselves against threat actors who are jailbreaking closed models for cyber attacks.
Model superiority
What did Mistral's CEO say?
Mistral's CEO Arthur Mensch claimed at a conference in Abu Dhabi that their new AI model outperforms Chinese models in some areas, including cybersecurity.
He said, "The model we're actually announcing today is actually above the Chinese models on certain aspects, including cybersecurity. So the narrative that Europe cannot compete is something that is not true."
However, he did not specify which Chinese models or how exactly Mistral's model surpassed them.