OpenAI has rolled out Ultrafast mode for GPT-6.1 Sol, the fastest tier in its API. The model generates up to 300 tokens per second: up to 8x faster than standard in Codex and up to 6x in the API. Speed costs 6 times more. Here's what GPT-6.1 Sol is, what Ultrafast costs and where it actually pays off.
What GPT-6.1 Sol is
GPT-6.1 Sol is OpenAI's upgraded mid-tier model, introduced in late September. In OpenAI's own tests it nearly matches the flagship GPT-6 Astra in coding, computer use and professional tasks while costing 5 times less than Astra. Note that "5 times cheaper" is compared with Astra, not with every model on the market, as some retellings claim.
- API price: $2 per 1M input tokens, $0.10 cached, $10 per 1M output. GPT-6 Astra is $10 / $1 / $50.
- Context: up to 1.05M tokens with up to 128K output, per this review.
- Availability: in the API as
gpt-6.1-sol, plus ChatGPT Work and Codex on Plus, Pro, Business, Enterprise and Edu plans.
What Ultrafast mode is
Ultrafast isn't a new model but a paid speed tier for the same GPT-6.1 Sol. Answer quality is the same, but it arrives several times faster, up to 300 tokens per second. OpenAI also has a Fast mode at double price; Ultrafast costs as much as three Fast.
- Price: 6x standard: $12 per 1M input tokens, $0.60 cached and $60 per 1M output.
- Long prompts cost more: if a prompt exceeds 272K tokens, the whole request is billed at long-context rates, $24 input and $90 output.
- In the API: enabled with
service_tier: "ultrafast"in the Responses API, including US and EU data residency. - In Codex and ChatGPT Work: only on the new Pro 500 plan at $500 a month and some enterprise and education contracts. In Enterprise an admin has to enable it first.
Who needs Ultrafast and who doesn't
Ultrafast pays off where waiting costs more than tokens: support and sales chatbots that must reply instantly, voice assistants, AI agents doing dozens of steps in a row, and developers in Codex who care about edit speed.
For bulk tasks it's a bad deal. Example: 1M output tokens cost $10 on the standard tier and $60 on Ultrafast. If it doesn't matter whether the answer takes 2 seconds or 10, you're just paying 6 times more.
What it means for media buyers
- Copy, hooks and creative translations don't need Ultrafast. Standard GPT-6.1 Sol or cheap models like Claude Haiku 5.5 are better value. See our guide to ad creatives for affiliate marketing for how to fit AI into creative work.
- Chatbots in Telegram or on landers that answer leads in real time: here speed affects conversion, so Ultrafast is worth testing.
- Automation with agents, such as launching and checking campaigns or analysing stats. If an agent takes many steps, a 6x speed-up noticeably cuts run time.
- The $500/month Pro 500 plan only makes sense for teams living in Codex all day. Solo buyers are better off paying per task via the API.
FAQ
How much does GPT-6.1 Sol Ultrafast cost?
6 times the standard tier: $12 per 1M input tokens and $60 per 1M output. In Codex and ChatGPT Work it's on the Pro 500 plan at $500 a month.
How much faster is Ultrafast?
Up to 300 tokens per second, up to 8x faster in Codex and up to 6x in the API, according to OpenAI. There are no independent measurements yet.
How is GPT-6.1 Sol different from GPT-6 Astra?
Sol is the cheaper model: in OpenAI's tests it's close to Astra in coding and office tasks but costs about 5 times less.
How do I enable Ultrafast in the API?
Set service_tier: "ultrafast" in a Responses API request for GPT-6.1 Sol.
Sources: OpenAI, VentureBeat, AlphaSignal, Kingy AI.