Cerebras Systems and Perplexity AI have partnered to create a new AI-powered search system called Sonar that processes 1,200 tokens per second. As reported by Michael Nuñez for VentureBeat, the system runs on Cerebras’s specialized AI chips and is built on Meta’s Llama 3.3 70B foundation model. According to Perplexity’s internal testing, Sonar achieves factuality scores of 85.1 out of 100, outperforming both GPT-4o mini and Claude 3.5 Haiku. The system will initially be available to Perplexity Pro users. Perplexity CTO Denis Yarats credits Cerebras’s AI inference infrastructure for enabling unprecedented speed and efficiency. The partnership comes shortly after Cerebras demonstrated its DeepSeek implementation, which showed speeds 57 times faster than traditional GPU-based solutions.