Cerebras has reportedly set a world record with its processing speed of over 2,500 trillion operations per second (T/s) on the 400 billion parameter Llama 4 Maverick model. This achievement has been met with skepticism from users who argue that the comparison should consider overall throughput rather than single-query performance. Concerns regarding the cost-effectiveness of Cerebras hardware have also been raised, particularly when comparing its performance to Nvidia's DGX B200, which significantly reduces costs for similar tasks. Critics point out that while Cerebras boasts impressive latencies, their throughput metrics are elusive, and the cost-performance ratio may not be favorable in a practical computing context. Additionally, the limitations of their SRAM-based architecture and reliance on massive investments for inference capabilities have been highlighted, leading to doubts about the long-term viability and scalability of their technology. There are calls for improved accessibility through an API and a push for improved model utility in their offerings.