OpenAI previews Ultrafast tier running GPT-5.6 Sol at 14x speed on Cerebras

Ultrafast is OpenAI selling latency, not capability. Running GPT-5.6 Sol at up to 14x standard speed — around 750 output tokens per second on Cerebras hardware — the tier targets real-time agents and voice applications where responsiveness, not raw intelligence, is the bottleneck. Hacker News (220 points, 75 comments) landed on exactly that read: 'latency, not capability, is the thing being sold.'

The business signals around it are arguably bigger than the product. CFO Dali Friar told investors that enterprise revenue has surpassed consumer — a milestone for a company built on ChatGPT's consumer virality, and evidence the monetization center of gravity is shifting to businesses paying for agents and API volume. Simultaneously, OpenAI scaled back NVIDIA funding guarantees for its Ohio data center from $250B to under $120B, a notable retrenchment amid the week's broader compute-financing frenzy.

The Cerebras partnership is strategically interesting: leaning on a non-NVIDIA accelerator for the speed tier, just as OpenAI trims its NVIDIA data-center commitments, hints at supply diversification. The reduced guarantee cuts against the maximalist capex narrative everyone else is pushing this week. Watch two things: whether enterprise-over-consumer revenue holds as a durable trend ahead of a rumored IPO, and whether Ultrafast's Cerebras dependency signals a real hardware hedge or just a niche latency product. Reddit also circulated an OpenAI-employee summary of upcoming ChatGPT speed improvements (490 upvotes), suggesting speed is a company-wide push.