Cerebras and Mistral: Revolutionizing AI Inference with Speed Record

Cerebras Systems, a pioneering AI chip manufacturer, has partnered with Mistral AI, a leading European AI startup, to achieve a remarkable speed record in AI inference. The collaboration has resulted in the integration of Cerebras' Wafer Scale Engine 3 (WSE-3) with Mistral's flagship 123B parameter model, enabling over 1,100 tokens per second on text queries. This breakthrough in AI performance is made possible by the WSE-3's SRAM-based inference architecture in combination with speculative decoding techniques developed in collaboration with researchers at Mistral.

The WSE-3 AI chip, built on TSMC's 5nm process, packs 4 trillion transistors, 900,000 AI-optimized compute cores, and delivers 125 petaFLOPS of peak AI performance. With 44GB of on-chip SRAM, the WSE-3 can store massive models and datasets, enabling faster and more efficient AI inference. The WSE-3's superior performance and power efficiency make it an ideal choice for training and deploying large-scale AI models.



Mistral AI's Le Chat platform, powered by Cerebras' WSE-3, offers instant responses to user queries, making it 10x faster than popular models such as ChatGPT 40, Sonnet 3.5, and DeepSeek R1. This significant improvement in speed is a testament to the power of the WSE-3 and the collaborative efforts of Cerebras and Mistral. The partnership between these two innovative companies is set to revolutionize the AI industry, pushing the boundaries of what's possible in AI inference and user experience.



In conclusion, the partnership between Cerebras Systems and Mistral AI has resulted in a remarkable speed record in AI inference, with the integrated WSE-3 and Mistral's flagship model achieving over 1,100 tokens per second on text queries. This breakthrough in AI performance is a testament to the power of the WSE-3 and the collaborative efforts of these two innovative companies. As the AI industry continues to evolve, partnerships like this will be crucial in driving innovation and pushing the boundaries of what's possible in AI inference and user experience.