"…model provider is unavailable"
The model behind the turn couldn't be reached. Pick a different model from the Model control and resend. If the turn was resumed from a checkpoint, the banner names the cause.
Rate limited or "try again"
Providers occasionally throttle. The agent retries automatically a few times; if the turn still fails, wait a minute and resend, or switch model family (Gemini ↔ Claude ↔ ChatGPT).
Premium models are greyed out
Premium models (the higher-cost tier) are not selectable on the Free plan. Upgrade, or use the standard model in the same family.
"Today's tokens are used up" / "Tokens exhausted"
You hit the daily cap or the monthly allowance. The message says when it refills. Options: wait, upgrade, or use Lite effort so the remaining credits go further. See Plans & credits.
The answer is truncated
Very long deliverables can hit output limits. Ask for the section that was cut ("continue the Discussion from where it stopped") — the artifact is extended as a revision. /compact frees context in long sessions.
The turn used far more credits than expected
Deep effort, premium models, video analysis and very broad research questions cost the most. The token count under each answer and Customize → Usage show where it went; narrow the question or lower the effort next time.