OpenAI sells GPT-6.1 Sol at up to 8x the speed for six times the price

OpenAI has started rolling out Ultrafast mode for GPT-6.1 Sol in the API, in Codex and in ChatGPT Work [1]. The company says it delivers near-Astra intelligence at up to 8x faster speeds than Sol Standard [1]. The catch is the price: in the API, Ultrafast costs $12.00 per million input tokens and $60.00 per million output tokens on short-context prompts, six times Sol’s Standard rate of $2.00 and $10.00 [2][5]. Speed is now something you buy per call. For anyone running agents or coding tools on OpenAI’s models, the question is no longer only which model to use but how fast you need each answer, and what that is worth.

What was announced

OpenAI posted the launch to its developer community and its developer account on 8 October [1][4]. Ultrafast is now rolling out for GPT-6.1 Sol in three places: the API, Codex and ChatGPT Work [1]. OpenAI describes it as built for work where speed and intelligence both matter, and names three examples: debugging an outage, agents navigating apps, and live experiences where every second counts [1].

The speed claim needs reading carefully. The announcement says “up to 8x faster speeds than Sol Standard” [1]. According to [4], OpenAI’s DevDay recap is more specific, giving up to 8x faster token generation in Codex and up to 6x in the API. RuntimeWire also points out that these are OpenAI’s stated maximums, not guaranteed response times for every request [4]. In other words, the API figure may be lower than the headline, and your own gain depends on your prompts, your tool calls and your network.

OpenAI’s Ultrafast guide calls it “the fastest service tier in the OpenAI API” and says to use it “when speed justifies the higher cost” [3]. It is broadly available for GPT-6 Astra and GPT-6.1 Sol, with preview access for GPT-5.6 Sol [3]. Developers select it by setting the model to gpt-6.1-sol and the service tier to ultrafast [3][5]. The guide strongly recommends WebSockets, especially for agentic applications that make many tool calls in quick succession, because without a persistent connection network overhead can reduce the latency gains [3]. HTTP requests through the SDK also work [3].

The prices are on OpenAI’s pricing page [2]. For short-context prompts, GPT-6.1 Sol costs $2.00 per million input tokens, $0.10 for cached input, $2.50 for cache writes and $10.00 for output at Standard. In Ultrafast mode the same four figures are $12.00, $0.60, $15.00 and $60.00 [2]. The model page states the rule plainly: Ultrafast prices are 6x Standard, and Fast mode prices are 2x Standard [5]. Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the whole request [5]; on the pricing page, that takes Ultrafast to $24.00 input and $90.00 output per million tokens [2]. Regional processing adds a 10% premium where it is available [5].

OpenAI frames the price against its bigger model. The announcement says Ultrafast Sol costs “just 1.2x the cost of Astra” [1]. The pricing page lists GPT-6 Astra at $10.00 input and $50.00 output at Standard, and Ultrafast Astra at $60.00 and $300.00 [2].

What changed, and for whom

The first group is developers paying for the API. Ultrafast for GPT-6 Astra and GPT-6.1 Sol is available to all API users [3]. It has its own rate limits, separate from Standard and Fast: for GPT-6.1 Sol, 1 million tokens per minute on the Build tier, 4 million on Launch and 40 million on Grow [3]. OpenAI tells organisations to check their limits before increasing traffic [3]. The practical change is that you can now choose speed per request. RuntimeWire notes that a team can send latency-sensitive requests to the six-times-priced tier and leave other traffic on standard service [4].

The second group is people who use Codex or ChatGPT Work through a subscription. Here access is narrower. OpenAI says Ultrafast in Codex and ChatGPT Work is available on Pro 500, eligible usage-based Enterprise and credit-based Edu plans, and that Enterprise admins must enable it [1]. If you are on another plan, the announcement does not offer you Ultrafast in those apps [1].

The third group is organisations with data-location rules. Ultrafast for GPT-6.1 Sol supports US and EU data residency as well as global processing [1][3]. GPT-6 Astra in Ultrafast supports US data residency and global processing only [3]. In the same announcement, OpenAI added EU data residency for GPT-6.1 Sol Fast and GPT-6 Luna Fast [1].

The comparison that matters for most buyers is with Fast mode, which already existed. Fast mode doubles Sol’s Standard price; Ultrafast multiplies it by six [5]. On the pricing page, Fast Sol costs $4.00 input and $20.00 output per million tokens, against $12.00 and $60.00 for Ultrafast [2]. OpenAI does not publish a speed figure for Fast mode in these sources, so the gap between the two tiers is something you would have to measure yourself.

Who it is for — and not

Ultrafast is for work where a delay has a real cost. OpenAI’s own examples are debugging an outage, agents moving through apps and live experiences [1]. A support agent that answers a customer in real time, a coding agent that a developer is watching, or an incident tool used while a service is down are all cases where faster output can be worth paying for. If someone is waiting on every token, the six-times price may be justified.

It is also for teams that already split traffic by urgency. The tier has its own rate limits and its own price, so it can sit beside Standard traffic rather than replacing it [3][4].

It is not for batch work. Summarising documents overnight, tagging records or running evaluations gains nothing from a faster answer, and OpenAI’s pricing page already offers Batch and Flex at 50% below the Standard price for that kind of job [2][5]. Paying 6x for speed nobody sees is the most likely way to waste money with this release.

It is not a new model. RuntimeWire puts it simply: Ultrafast keeps the same model and sells a shorter wait for its output [4]. If Sol’s answers are not good enough for a task at Standard, they will not be better in Ultrafast.

It is not available to every subscriber. Inside Codex and ChatGPT Work it is limited to specific plans, and Enterprise admins must turn it on [1]. Solo users on other ChatGPT plans can still reach it through the API, at API prices [3].

And it is not a promise of a fixed speed. “Up to 8x” is a ceiling, and according to [4] the API figure OpenAI gives in its DevDay recap is up to 6x. Test before you budget.

This page will be updated if OpenAI changes Ultrafast prices or plan access, publishes measured speeds for the API, or extends the tier to more models.

∴ Speed is now a line item: pay six times Sol’s standard rate only on calls where waiting costs you something.