Archive — US — techcrunch
Mistral’s new 1T model aims to leapfrog closed and open rivals
Mistral AI has launched ML4, a new 1 trillion-parameter multimodal model aiming to outpace both American and Chinese AI rivals. French President Macron promotes it as a 'third way in AI.' While currently accessible via a public endpoint, Mistral plans to release open weights soon. ML4 is designed to excel in cybersecurity and finance, leveraging fewer resources than competitors, backed by major investors.
Key points
- Mistral AI introduced ML4 as a 1 trillion-parameter multimodal model.
- ML4 aims to provide an alternative to both closed and open AI models.
- Currently, ML4 is not an open-weight model, accessible only through a public guardrail endpoint.
- Mistral plans to release open weights within three weeks after safety testing.
- ML4 was trained on 4,000 Nvidia GPUs, significantly less than competitors.
- Optimized for use cases in cybersecurity, finance, and chip design.
- Mistral is backed by major firms like ASML and Samsung, positioning itself as an innovative AI lab.
Why it matters
- Mistral's ML4 reflects the broader trend of seeking balanced AI models amidst escalating geopolitical tensions over technology.
- The debate surrounding open versus closed AI models is central to current discussions on AI ethics and security.
- Mistral's development aligns with the focus on resource-efficient AI training, which is becoming increasingly important in 2025 IT trends as companies seek to reduce costs and environmental impact.
Mistral's ML4 represents a strategic innovation in the AI landscape, aiming to bridge the gap between open and closed models while catering to security-conscious enterprises, positioning the company as a competitive player in the evolving AI market.