With the release of its new trillion-parameter model, Mistral is hoping to demonstrate it’s “still in the race” to build frontier-level artificial intelligence.
The Paris-based startup, which has quickly become a major player in the open-source AI world, announced the model, dubbed “Le Chonk,” on Monday. It’s the latest in a string of increasingly powerful open-weight models released by Mistral, and the company claims it outperforms other open-source options—and even some closed-source models—on a variety of benchmarks.
“We wanted to show that we are still in the race, and that we can still deliver state-of-the-art models,” says Mistral CEO Arthur Mensch in a press briefing. “We think this is a very important step forward for open-source AI.”
Le Chonk is a mixture-of-experts model, meaning it’s composed of multiple smaller models that are activated depending on the input. This allows it to achieve high performance without requiring as much computational power as a traditional dense model. The model has 1.1 trillion parameters, and Mistral claims it outperforms Llama 2 70B on most benchmarks.
“It’s a really impressive model,” says Miles Brundage, a researcher at the OpenAI safety team who has been following the development of open-source AI. “It’s a sign that the open-source community is catching up to the closed-source leaders.”
Mistral is releasing Le Chonk under the Apache 2.0 license, which means it can be used for commercial purposes without restriction. The company is also making the model weights available for download, allowing anyone to run it on their own hardware.
The release of Le Chonk comes at a time when the AI landscape is rapidly evolving. Companies like OpenAI, Google, and Anthropic are all racing to build ever-more-powerful AI models, and the competition is fierce. Mistral’s success in the open-source world is a testament to the power of collaboration and the growing demand for accessible AI technology.