GPT-5.6 Sol Ultrafast: OpenAI hits 750 tokens/sec on Cerebras
Frontier Sol now runs 14x faster with no quality drop—but behind a gated preview with no price yet. Cursor's cloud agents also boot 3x faster, free.
Copy markdown
750 tokens/sec, and no quality tax
Cerebras' wafer-scale engine runs GPT-5.6 Sol at up to 750 output tokens/sec—14x faster than the standard tier and 5x faster than Opus 4.8 on Fast mode. OpenAI reports identical GDP-Val scores and a 5.6x end-to-end speedup with no measured quality loss.
The catch: a gated preview, no price
Ultrafast is limited to “a select group of customers” through the OpenAI API, with no GA date and no pricing announced. You can request access, but unless you're on the list you can't build on it today.
What real-time frontier speed unlocks
This is frontier-grade intelligence at conversational speed: sub-second incident triage, live fraud checks, and overnight batch research turned into interactive loops. It cleared 2,500 Humanity's Last Exam questions in 11 hours versus 78 for Claude Fable 5.
Cursor cloud agents boot 3x faster
Cursor's new “builds” pre-bake your repo, dependencies, and install scripts into cloud-agent environments—cutting time-to-first-token 3x at no extra cost, and keeping the last working build when a fresh commit breaks the setup.