1 min read

Mistral AI Releases Large 4 With One Trillion Parameters

The model was trained using 4,000 NVIDIA GPUs. Mistral plans to release its weights after safety testing in about three weeks.

A graphics processor mounted in a laboratory test rig / TokenPost.ai (mono)
A graphics processor mounted in a laboratory test rig / TokenPost.ai (mono)

Mistral AI released Mistral Large 4 on Oct. 6, introducing a one-trillion-parameter model aimed at finance and other specialized uses as the French company seeks a place among leading artificial intelligence developers.

The model was trained using 4,000 NVIDIA GPUs, and Mistral plans to make its weights available in about three weeks after safety testing. Until then, access is limited to a public endpoint with safeguards.

Mistral Large 4 is a multimodal model, meaning it can work with more than one type of input. Its intended applications include cybersecurity, finance and chip design.

The model’s weights are not yet openly available. Mistral plans to work with selected partners and governments before releasing them, with safety testing to be completed first.

Mistral Large 4’s performance against other models has not yet been established. The company has positioned it as a potential competitor to both open-weight and closed models, but benchmark results are still pending.

Simon Yoon

Reporter

Simon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.

Loading…