OPENAI’S MOST POWERFUL NEW MODEL JUST BECAME 10 TIMES FASTER — 750 TOKENS PER SECOND ON CEREBRAS HARDWARE Speed was always the missing ingredient for AI agents. You can build the most intelligent model on the planet, but if it takes 30 seconds to respond, people stop using it. OpenAI just solved that problem. The company announced it is launching GPT-5.6 Sol, its most capable model in the new 5.6 series, on Cerebras wafer-scale hardware at up to 750 tokens per second. The standard GPU infrastructure serving frontier models today runs at roughly 40 to 120 tokens per second. What Cerebras delivers is not an incremental improvement. It is a different category. A task that takes a standard cluster 30 seconds now takes under three. For AI agents that need to reason through long chains of steps, write code, test it, find the errors, and iterate, the difference between 30 seconds and 3 seconds is the difference between something people will actually use and something they will abandon. GPT-5.6 Sol is already the highest-scoring model ever recorded on Terminal-Bench 2.1, the test that measures complex command-line AI workflows. It is also OpenAI’s strongest model to date for vulnerability research and security tasks. Broad access remains limited while the US government finishes its safety review, but the Cerebras partnership makes clear that the next phase of the AI race will be fought on inference speed, not model size alone. Keywords: GPT-5.6 Sol Cerebras, OpenAI inference speed, AI tokens per second, frontier model speed