OpenAI releases GPT-6.1 Sol Ultrafast on day 4 of its 28-day marathon

On day four of its "28 days in a row: release or reset" marathon, OpenAI introduced GPT-6.1 Sol in an Ultrafast version. The day before, on day three, the company opened GPT-6 to all users, free and paid. The new model arrived with an updated control system that reacts faster to instruction changes made mid-task. The series has 24 days left to run.

What we know about the release

The source, the Chinese outlet Jiqizhixin, gives a brief picture. GPT-6.1 Sol comes in an Ultrafast variant, meaning it prioritizes response speed. The second part of the release is an updated model control system. According to the description, it lets the model adapt more quickly when the user changes the conditions of a task on the fly. For long sessions and agentic scenarios this matters: an edit made mid-execution should not be wiped out or ignored.

Keep in mind how this is sourced. It is a retelling of an announcement, not a technical report. The material contains no concrete figures for latency, throughput, price per token or context size. We don't have them either yet, so any comparison with previous versions would be guesswork.

The marathon as a format

The "28 days in a row" series is tied to Tibo, whom Chinese colleagues call "大神". The format is simple: every day brings either a new release or a reset of something that already exists. The word "重置" in the original allows several readings; it probably refers to resetting limits or service state. The logic is clear enough, though: a continuous stream of events holds developers' attention and keeps competitors from taking over the news cycle.

The sequence of the first days shows the rollout order. First, the base GPT-6 model goes out to every plan. Then version 6.1 arrives with a accelerated mode. Free users get the flagship generation, while paid users and API customers get separate speed and quality profiles.

What is confirmed and what is not

• Model: GPT-6.1 Sol, Ultrafast variant.
• Announcement date: October 9, 2026.
• GPT-6 availability: all users, free and paid, since the previous day.
• New in control: fast reaction to changes in instructions.
• Position in the series: day 4 of 28.
• Not disclosed: latency, tokens per second, price, context window, benchmark results.
• Not disclosed: Ultrafast availability by plan and in the API.

What teams should do

It is too early to switch to Ultrafast on the strength of one announcement. A more sensible approach is to assemble your own set of 20–50 real queries and run it through both GPT-6 and GPT-6.1 Sol Ultrafast. Measure time to first token, total response time and the share of tasks solved without a repeat correction.

Separately, test controllability. Give the model a multi-step task, change a condition at step two and see whether the context holds. This is the scenario where the claimed improvement should show up.

If you build on the API, it is better to wait for the official price and limits table. Fast modes usually cost more than base ones, and the payoff depends on where latency is critical in your product: chat, code autocompletion or a voice interface. The remaining 24 days of the marathon may bring follow-ups, so keep your architecture flexible, with the ability to swap models quickly.