"Selected model is at capacity" in Codex: what to do while OpenAI hasn't fixed it

The error "Selected model is at capacity. Please try a different model." started showing up in Codex right after OpenAI launched GPT-6 Astra, in early September 2026. Switching models usually doesn't fix it, and it's not your quota running out: it's OpenAI's server rejecting the session. Here's how to confirm that before wasting time, and how to set up a fallback provider so you're not stuck depending on a single engine.

What does "Selected model is at capacity" mean?

It's a server-side error on OpenAI's end, not your quota. Issue #43738, opened on 2026-09-08 in the official Codex repository, collects the reports: a recurring error on Astra and on older Sol models, across CLI, IDE extension and the web version, with unavailability windows lasting more than 5 hours. One report describes a roughly 2-hour task that burned through 80% of the weekly quota before hitting this error, without finishing the work.

Is it my quota or is it the server?

Run /status to see the session configuration and token usage, and /usage to see account activity in a daily, weekly or cumulative view. If your quota is nowhere near the limit and the error persists, the problem isn't on your end.

Also check status.openai.com before trying command after command. A similar incident already happened, declared and resolved by OpenAI itself on 2026-06-16: about 3 hours, classified as a performance degradation across Codex components.

Does switching models fix it?

Usually not. In the reports gathered in issue #43738, switching models hits the same error most of the time, because the cause is admission and routing on the server, not the specific model you picked. If the thread still holds the conversation state, the next request that goes through resumes the work where it left off. That's the only documented way out besides waiting.

Can you set up a fallback provider in Codex?

Yes. Codex's config.toml accepts multiple model_providers, and switching engines doesn't require editing the file every time the error shows up.

[model_providers.verboo]
name = "Verboo Code"
base_url = "https://code.verboo.ai/router/v1"
env_key = "VERBOO_API_KEY"
wire_api = "chat"

With the key exported (generated at code.verboo.ai/api-keys):

export VERBOO_API_KEY="your-key-here"
codex --config model_provider='"verboo"' --model glm-5.2

That switches the engine for just that session. To make it reusable, save it as a profile at ~/.codex/verboo.config.toml and call it with codex --profile verboo whenever "at capacity" shows up again.