OpenAI says its faster AI tool cut engineers' outage checks to minutes

OpenAI says its engineers now investigate technical incidents, such as outages, in minutes instead of hours. The company credits a faster processing setting called Ultrafast. The claim comes from OpenAI itself, and no independent test has backed it up.

What OpenAI said

In an account published October 26, 2026, OpenAI said engineers use Ultrafast to investigate incidents in minutes. They also use it for security analysis, refactoring (cleaning up code without changing what it does) and daily tasks. OpenAI framed this as part of a broader push to speed up engineering work, not as a product built for incident response.

Ultrafast is not a separate product. OpenAI describes it as a lower-latency processing tier. Latency is the wait for an answer, so this is a faster way to run its models, not a new tool.

What the earlier preview showed

An August 13, 2026 preview described how OpenAI uses it for incidents. The workflow reads application logs, analyzes traces and recent code changes, and pulls together engineer conversations. It also suggests follow-up checks and helps prepare or validate fixes. OpenAI said engineers remain responsible for judgment and deployment.

That preview cited speeds of "up to 14 times" and 750 output tokens per second. Those figures applied only to GPT-5.6 Sol, not to every current model. Newer background says GPT-6 Astra Ultrafast can generate tokens up to eight times faster than Astra Standard. A token is a small chunk of text, so that measures how fast text is produced, not how fast a problem gets solved.

Who can use it

Ultrafast is available on Pro 500 and eligible Enterprise and Edu plans. Enterprise access requires workspace permission. Current documentation lists GPT-6 Astra and GPT-6.1 Sol support in Codex and ChatGPT Work.

What is not known

• No independent before-and-after timing study of incident handling has been provided.
• The hours-to-minutes result is OpenAI's internal account, not a universal benchmark.
• The eightfold figure is about text speed, not time to resolve an incident.

For teams that qualify, the practical step is to time a few real tasks with and without Ultrafast. Keep a person in charge of final decisions, as OpenAI says its own engineers do.