OpenAI Launches Sol Fast, a 750-Tokens-per-Second Tier of GPT-5.6 on Cerebras
OpenAI has begun serving GPT-5.6 Sol on Cerebras wafer-scale chips at up to 750 tokens per second, about 10x typical Nvidia throughput for a frontier model, under a new premium tier called Sol Fast priced well above standard Sol.
OpenAI has started routing GPT-5.6 Sol traffic through Cerebras wafer-scale hardware this week, marking the first OpenAI frontier model deployed at production scale on non-Nvidia silicon. The new premium tier, Sol Fast, targets latency-sensitive agentic workloads where sequential tool calls otherwise stack up wait time.
- Sol Fast: up to 750 tokens/second, priced at $12.50 / $75 per million input/output tokens
- Standard Sol: $5 / $30 per million tokens, running on Nvidia clusters at roughly 70 tokens/second
- Each transformer layer runs on a dedicated Cerebras WSE-3 wafer, spanning an estimated 70 to 100 wafers per deployment
- Access is limited to select customers at launch, with OpenAI expanding capacity through July
The deal follows a multi-year OpenAI-Cerebras agreement signed in January 2026 for 750 megawatts of wafer-scale inference capacity, and lands weeks after Cerebras' May IPO. Compare Sol Fast against other coding-oriented models in the arena.