OpenAI Deploys GPT-5.6 Sol on Cerebras at Up to 750 Tokens Per Second
Summary: OpenAI confirmed it will run GPT-5.6 Sol on Cerebras wafer-scale hardware in July at up to 750 tokens per second — roughly an order of magnitude faster than typical GPU-based frontier inference.
Key Facts
- Cerebras throughput: up to 750 t/s for Sol; standard GPU clusters serve frontier models at 40–120 t/s
- Pricing: Sol at $5 input / $30 output; Terra at $2.50/$15; Luna at $1/$6 (per 1M tokens)
- Rollout limited to ~20 select partners initially, following the White House's voluntary pre-deployment safety review request
- Sol targets agentic, long-horizon, and scientific reasoning tasks; Terra and Luna are optimized for cost-sensitive general workloads
Why It Matters
Serving a frontier model at 750 t/s collapses multi-step agent loops from overnight jobs to real-time interactions. For developers building coding assistants, research agents, and complex orchestration pipelines, this speed tier removes the biggest practical bottleneck — waiting for the model — and resets expectations for what interactive AI tooling can do.