OpenAI Partners With Cerebras to Run GPT-5.6 at Record Inference Speed

OpenAI has partnered with AI chip startup Cerebras to run inference for its next-generation frontier model, GPT-5.6 Sol, on the company's computing platform. Generation speed for large models directly affects service latency and costs, making the collaboration an important technical test of Cerebras' ability to handle flagship-model workloads.

The latest tests showed GPT-5.6 Sol reaching an inference speed of 750 tokens per second. Analyst Serenity said she had established a position at about $170 per share because OpenAI's adoption validated Cerebras' technology, though she cautioned that its valuation remained high. As of July 20, 2026, the provided information did not specify the date of a formal announcement.