Mistral Unveils First Reasoning Model to Rival OpenAI and DeepSeek

French AI company Mistral, supported by Microsoft, is set to launch its inaugural reasoning model on Tuesday, positioning it as a competitor to offerings from OpenAI and China’s DeepSeek.

During a fireside chat at London Tech Week, Mistral CEO Arthur Mensch told CNBC’s Arjun Kharpal, “We’re rolling out our new reasoning model in a few hours. It’s highly competitive with others and uniquely capable of reasoning in multiple languages.”

PARIS, FRANCE - MAY 22: French founder of artificial intelligence start-up Mistral AI, Arthur Mensch attends the Viva Technology show at Parc des Expositions Porte de Versailles on May 22, 2024 in Paris, France. Viva Technology, the biggest tech show in Europe but also in a unique digital format, for 4 days of reconnection and relaunch thanks to innovation. The event brings together startups, CEOs, investors, tech leaders and all of the digital transformation players who are shaping the future of the Intern
French founder of artificial intelligence startup Mistral AI, Arthur Mensch, attends the Viva Technology show at Parc des Expositions Porte de Versailles in Paris, France, on May 22, 2024.

Reasoning models are designed to tackle complex tasks through step-by-step logical processes. Mensch highlighted that Mistral’s model excels in mathematics and coding.

Current reasoning models on the market include OpenAI’s o1, integrated into ChatGPT last year, and DeepSeek’s R1. Mistral, known for its open-weight large language models, makes its model parameters publicly available. This allows developers to access and tweak the model’s core knowledge, reducing the time and cost of building systems from the ground up.

Mensch emphasized that Mistral’s Magistral reasoning model stands out for its ability to reason in European languages. “While U.S. models typically reason in English and Chinese models in Chinese, our model focuses on European languages,” he said, noting plans to support additional languages in the future.

Earlier this year, DeepSeek’s R1 model made waves in the AI community and global markets, offering performance comparable to OpenAI’s o1 at a lower cost.