✳ the wire · analysis
"Selected model is at capacity": Codex and ChatGPT turn users away from GPT-6.1 Sol and GPT-6 Astra while the API still answers
Get launch alerts like this, free →
Since early on October 9 (UTC), Codex and ChatGPT users have been posting the same error when they pick GPT-6.1 Sol or GPT-6 Astra: "Selected model is at capacity. Please try a different model." Some say it stops agent tasks halfway through. OpenAI's status page showed no incident, and the API kept answering.
WHAT IS SOURCED
The error and its timing come from users on X: more than a dozen accounts, posting in English and Chinese between 00:33 and 08:45 UTC, naming Codex, ChatGPT, Sol and Astra. One dated it to the afternoon of October 8, US time. These are user reports, not measurements.
OpenAI's status page, read at 08:43 UTC, said "All Systems Operational". Its most recent incident, errors in ChatGPT Work for some APAC users, was marked resolved on October 8.
The API side we measured. ArtificialWatch's own request to GPT-6.1 Sol answered in 0.4 seconds, and GPT-6 Astra answered too (our status page, checked within the hour before 08:50 UTC). OpenRouter's endpoint stats, read at 08:47 UTC, put OpenAI's standard GPT-6.1 Sol endpoint at 100% uptime over the past day and its Flex endpoint at 55.6%. Flex is OpenAI's discounted tier, and OpenAI's docs say it "may sometimes lack sufficient resources" and then returns a 429 Resource Unavailable error.
THE BACKDROP
Two speed changes landed on these models this week. On October 5, OpenAI's Thibault Sottiaux (@thsottiaux) posted that default speed was about 50% faster across GPT-6 Astra and GPT-6.1 Sol on subscriptions, including apps that sign in with ChatGPT. On October 8, OpenAI rolled out Ultrafast for GPT-6.1 Sol in the API, Codex and ChatGPT Work. We do not know whether either is connected to the errors.
WHAT WE DO NOT KNOW
The cause. OpenAI had not addressed it on its status page or from its developer accounts on X as of 08:50 UTC.
Whether Sol has gotten worse. Several posts say it feels dumber, but nobody has published the same prompts run before and after. A capacity error is a load problem, which is a different thing from a quality drop.
How long it lasts. Our status page checks every model it tracks by sending a real request: artificialwatch.com/status.
Source: Users on X (reports) · OpenAI status page · ArtificialWatch's own model check · OpenRouter endpoint stats · OpenAI's Flex processing docs ↗ · GPT-6.1 Sol tracker · the bench index
