719 hours, 50 minutes. That's what Claude's extended thinking display reportedly showed after processing one user's prompt - roughly 30 days of continuous reasoning. The result? Zero sources found.
Claude's extended thinking feature works by letting the model reason through problems step by step before returning an answer, similar to how a person might work through a hard math problem on scratch paper before writing the final answer. The feature normally shows how long that reasoning process took, usually measured in seconds or minutes. A display reading 719h 50m is almost certainly a timestamp calculation bug, not evidence that Anthropic's servers spent a month chewing on a single query.
But the zero-sources result is the part worth sitting with. Extended thinking is supposed to make Claude better at hard research and reasoning tasks. Getting a maximally absurd thinking duration paired with a completely empty result is a sharp illustration of something Claude users hit periodically: the model can spin its wheels with great confidence and deliver nothing useful. Anthropic hasn't commented on this specific incident. The bug appears isolated - no widespread reports of similar timestamp errors have surfaced.