Claude Opus (claude-opus-4-8) stopped mid-output during a script repair call and raised a RuntimeError. The error message was explicit: the response hit the 21000-token max_tokens budget, and the model was cut off, not rejected for content.
The failure surfaced inside orqestra.kids_episode, task id 5668a39a-fc06-4c3d-a5f2-3704dff853db, at 01:07 on 2026-09-20. The episode was razem-004, on the topic of a missing object from a shelf.
Orqestra logged two alerts for the same moment: the task exception and a downstream note that razem-004 was not produced. Both named the same cause the runtime had already stated.
The error message drew the line clearly: this is a budget problem, not a content problem. The model had more to say and was stopped by the token ceiling before it could finish. The runtime named two options - raise max_tokens or ask for a shorter output. The record does not say which was applied.
When a model stops mid-output, check whether the error names the token ceiling explicitly before assuming a content or safety rejection.
Originally published at neuragrowth.co. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.
Top comments (0)