What happened
OpenAI is previewing Ultrafast mode, powered by Cerebras infrastructure from a ten-billion-dollar partnership signed earlier this year. The service delivers up to 750 output tokens per second from GPT-5.6 Sol and will initially be available only through the OpenAI API to select customers, with gradual expansion as capacity grows.
Why it matters
The speed boost—described by OpenAI as combining smaller models' speed with large models' full capabilities—unlocks use cases that require real-time analysis. OpenAI highlights incident response (analyzing logs and code during outages), finance (evaluating market signals live), customer support (resolving complex inquiries instantly), and research (turning overnight batch jobs into interactive sessions). The company is already using the model internally for incident response.
What to watch
Ultrafast is the third tier in OpenAI's speed-based pricing ladder, following Fast Mode (up to 2.5x speed at roughly double the standard price). Companies interested in early access can sign up through a form on OpenAI's site.
Summaries like this, in your inbox every morning.
OpenAI's Ultrafast mode marks a strategic shift in how the company monetizes its AI output. Rather than charging solely on token volume or model capability, OpenAI is now selling speed itself as a tiered service—a model long established in cloud infrastructure (AWS charges premium rates for higher performance) but novel in the AI inference market. The ten-billion-dollar Cerebras partnership, announced earlier in the year, now delivers tangible product value: the ability to serve 750 tokens per second, positioning speed as a direct revenue lever. This tiered approach (standard, Fast Mode at 2.5x with roughly double pricing, and now Ultrafast at an implied premium) assumes that speed will become a binding constraint for many use cases, particularly where real-time decisions matter—trading, incident response, or customer-facing interactions. OpenAI's internal use of the model for incident response during outages suggests the company views this not just as a commercial offering but as architecturally sound for high-stakes workflows.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Vinci4D provides chip-design simulation software that trains on physics equations instead of customer data, an…
SEMI President and CEO Ajit Manocha told DIGITIMES that the semiconductor industry reached the US$1 trillion r…
Nvidia's AI Enterprise lists at $4,500 per GPU annually, and paid subscriptions add supported software, securi…
CoreWeave fell more than 25% this week, its debt yield briefly rose above 16% on Thursday, and SK Hynix droppe…
China's commerce and science ministries released a draft update adding 23 key AI technologies to its export co…
Anthropic told investors this week that it is making 85% gross margins, a figure the columnist calls a cash co…
