French company Mistral has unveiled a new open-source artificial intelligence model, ML4, which it claims is one of the best open-source models globally. This launch is part of a broader competition among Western companies to bridge the gap with Chinese developers in the AI market. ML4 boasts a trillion parameters and is highly efficient in tasks related to cybersecurity, programming, manufacturing, finance, and multimedia.
Mistral has long been considered Europe's sole response to OpenAI and Anthropic, and it has secured significant funding to support its endeavors. In September, the company raised €3 billion in a Series D funding round, which valued it at €21 billion. This substantial funding has solidified Mistral's position as one of the most well-funded and valuable AI startups in the region. The company's focus on open-source AI models has garnered attention, particularly as access to these models has become a point of contention in the technological rivalry between the US and China.
The launch of ML4 marks a significant milestone for Mistral, as it aims to provide an alternative to closed models developed by OpenAI and Anthropic. Open-source models like ML4 can be modified and hosted independently, offering greater flexibility compared to closed systems. Mistral has made ML4 available on a trial basis to developers, cybersecurity leaders, and government entities, with a broader release planned later this month. This strategic move is expected to enhance the company's position in the competitive AI landscape.
Training ML4 required 4,000 NVIDIA Grace Blackwell processing units over two months, utilizing Mistral's private data centers in Europe. According to the company, ML4 will be classified among the top open-weight models globally in terms of overall benchmark performance upon release. Furthermore, Mistral asserts that ML4 represents the strongest open-weight model developed outside of China, with a significant margin. However, the model still lags behind the latest technologies in certain areas, such as programming.
Guillaume Lamble, co-founder and chief scientist at Mistral, emphasized that ML4's capabilities will continue to improve as the company expands its training capacities. He also highlighted the model's cyber defense features, which will enable institutions and governments to protect themselves against threat actors that exploit closed model restrictions to launch cyberattacks. This focus on cybersecurity is particularly relevant, given the increasing use of AI models in various industries.
The emergence of open-source models like ML4 has significant implications for the AI landscape. Earlier this year, the open-source model GLM 5.2 from Chinese company Z.ai was used to counter an attack by rogue agents from OpenAI. This incident demonstrated the potential benefits of open-source models in enhancing cybersecurity and flexibility. As the AI market continues to evolve, the competition between open-source and closed models is likely to intensify.
The AI startup Reflection AI, backed by NVIDIA, has also entered the open-source AI market with its own model, claiming it rivals GLM 5.2 in capabilities. However, GLM 5.2 was surpassed by GLM 5.3 in August. The ongoing developments in the AI sector underscore the rapid progress being made in AI research and the growing importance of open-source models in driving innovation and competition.
Key points
- Mistral's ML4 model is one of the best open-source AI models globally.
- The model has a trillion parameters and is highly efficient in various tasks.
- The launch of ML4 marks a significant milestone in the competition between open-source and closed AI models.
