GPT-6.1 Sol Ultrafast tested: four times faster, six times the price
We ran GPT-6.1 Sol in Standard, Fast and Ultrafast on our secret coding test. Ultrafast was about 4x faster with the same perfect score, but a job cost 7.77 cents instead of 1.25 cents.
OpenAI has a new speed setting for its coding model GPT-6.1 Sol: Ultrafast. It writes answers much faster, but every token costs six times as much. Tokens are the small pieces of text an AI reads and writes. We gave GPT-6.1 Sol the same three coding jobs in all three speed settings. Ultrafast finished a job in 5.7 seconds, about 4 times faster than Standard. The answers were just as good: every secret test passed. But one job cost 7.77 cents instead of 1.25 cents.
What is Ultrafast?
GPT-6.1 Sol is one of OpenAI's newest models for coding and hard work. Programs can use it through OpenAI's API, the connection that programs use to talk to the model. Since 8 October 2026, you can add one setting to a request: service_tier: "ultrafast". The model stays the same, but OpenAI writes the answer faster. It is for all API users, OpenAI's changelog says.
An overview on OpenAI's forum, kept by a forum member, says Ultrafast is "up to 8x the speed" of Standard. It also says Ultrafast is coming to Codex and ChatGPT Work. We tested the API only.
Four speeds, four prices
OpenAI now sells GPT-6.1 Sol in several ways. The same answer costs a different amount depending on how fast you want it:
- Batch and Flex: half price, but you wait longer.
- Standard: the normal price.
- Fast: twice the normal price.
- Ultrafast: six times the normal price.
That comes straight from OpenAI's model page. So Ultrafast is the most expensive way to use this model. The question is: how much faster do you get for that?
What does Ultrafast cost?
For short questions, OpenAI's price list gives these prices per million tokens for GPT-6.1 Sol:
- Standard: $2 to read, $10 to write.
- Fast: $4 to read, $20 to write.
- Ultrafast: $12 to read, $60 to write.
For very long questions, over 272,000 tokens, Ultrafast costs $24 to read and $90 to write. With coding work, most of the money goes to writing: the code and the "thinking" the model does first.
Our test: same model, three speeds
We used the jobs from our private coding test, the same ones as in our Claude Haiku 5.5 test. The tests in our public test kit are known by now, so these stay secret. There are three jobs:
- D. Fix a money reader: code that turns "€ 1.234,56" into cents, with several bugs. 28 secret checks.
- E. Split a restaurant bill: shared dishes and a tip, with cents that must add up exactly. 17 checks.
- F. Read a CSV file: a spreadsheet saved as text, where commas, quotes and line breaks can hide inside a field. 17 checks.
We sent each job three times to GPT-6.1 Sol in Standard, three times in Fast and three times in Ultrafast. Each try was one request, with OpenAI's default thinking setting. Only the speed setting changed. In total that was 27 jobs.
The results: same answers, much faster
All three settings passed every secret test: 186 of 186. So Ultrafast did not make the model sloppier.
The difference was time. In the middle, one job took:
- Standard: 22.0 seconds, writing about 53 tokens per second.
- Fast: 15.1 seconds, about 72 tokens per second.
- Ultrafast: 5.7 seconds, about 206 tokens per second.
So Ultrafast was about 4 times faster than Standard in our test. That is a lot, but less than the "up to 8x" in the forum overview. Our jobs are short, so the start of the answer weighs heavily; with longer answers the difference may be bigger.
What the extra speed costs
One job cost us, on average:
- Standard: 1.25 cents.
- Fast: 2.37 cents.
- Ultrafast: 7.77 cents.
So Ultrafast was about six times as expensive per job, just as OpenAI's prices say. Compared with Standard, every second you save with Ultrafast costs about 0.4 cents. Fast is the cheaper middle way: about 7 seconds faster for about 1 cent extra.
For one job that is little money. But an app that sends 10,000 jobs a day would pay about $650 more per day for Ultrafast than for Standard.
Who is Ultrafast for?
Ultrafast makes sense when a person is waiting for the answer and every second counts. Think of a coding helper in an editor, a chat that writes code live, or an AI agent that has to take many steps one after the other. There, 6 seconds instead of 22 feels very different.
For work that runs in the background, like checking code at night, Standard or even Batch is much cheaper. And if you want speed but not the price, Fast may be enough. Earlier we compared GPT-6.1 Sol with GPT-6 Astra on the same kind of work.
How we checked
On 9 October 2026 we read OpenAI's changelog, the model page for GPT-6.1 Sol and the price list. The screenshots are our own captures of those pages, cropped only.
We ran the test the same day through OpenAI's Responses API with our own API key: 3 jobs, 3 tries, 3 speed settings, so 27 requests. OpenAI confirmed the speed setting in every answer. The costs are worked out from the token counts OpenAI returned and the list prices; the whole test cost about $1. The owner of this site paid for it. The charts were made by us. None of the images was made by AI.
This was a small test with three short jobs. Speeds can change from day to day, for example when many people use the service at the same time.