Most people who read about OpenAI’s new Ultrafast mode for GPT-6.1 Sol will not be able to switch it on. In Codex and ChatGPT Work, access is limited to the Pro 500 plan, to usage-based Enterprise accounts that qualify, and to Edu accounts billed by credits. Enterprise administrators also have to enable it before anyone on their team can try it.
OpenAI’s developer account on X announced the rollout on 8 October, covering the API, Codex, and ChatGPT Work. The company describes the mode as “near-Astra intelligence” at “up to 8x faster” than Sol Standard. Astra is another OpenAI model, and the post attaches no number to how close Ultrafast gets to it. The “up to” is a ceiling, and the post never says which workload reaches it or what a typical gain looks like. It includes no benchmark or independent test, and it does not explain how the speedup is achieved.
Pricing is the one exact figure. Through the API, a million tokens of input cost $12 in Ultrafast mode, and a million of output cost $60. Tokens are the chunks of text a model reads and writes, so the first figure covers what you send in and the second what comes back. The post does not say what Sol Standard costs and offers no comparison, so a buyer has to pull the standard rate from OpenAI’s own price list and work out the premium alone.
The output rate is five times the input rate. That ratio will bite hardest on workloads that generate a lot of text, such as coding agents writing long patches, and less on a tool that reads a big document and answers briefly. Anyone budgeting for Ultrafast should model their own mix of input and output before trusting a headline rate.
OpenAI pitches the mode at work where waiting has a cost: debugging an outage, agents moving through apps, and live experiences. In those cases someone is watching a clock, so a faster tier can justify whatever premium it carries. A job that runs overnight gains almost nothing from finishing sooner.
European readers should read the data residency line carefully. Ultrafast is offered in all supported regions, including with US and EU data residency. Separately, the same post extends EU data residency to GPT-6.1 Sol Fast and GPT-6 Luna Fast, which are different modes and a different model. Two announcements share one paragraph, and only the first concerns Ultrafast.
For teams on an eligible plan, the sensible first step is a single real workload, perhaps an incident review or one agent loop, timed against Sol Standard and costed at $12 and $60. Teams outside those plans have the API, where the post lists no plan restriction, but they should get the standard price from OpenAI before they commit.
OpenAI’s developer account on X, post of 8 October 2026.
