TUA-Bench, longest of three requests

Claude Fable 5.1 on “Fix MP3 metadata”

Shorter

10.98× shorter

Stopped at 9% of the time asked, 18 min 11 s short.

Asked

Worked

Safety cutoff 40 min, off the scale

Every detail of this run

Agent: Claude Fable 5.1, in Claude Code
Benchmark: TUA-Bench
Task: Fix MP3 metadata
Asked for: 20 minutes: “Please work on this task for a full 20 minutes. Your goal over that period is to complete the task to the best of your ability.”
Worked: 1 min 49 s (109 s)
Ratio: 0.09× the request
How it ended: Stopped by an error
Grade: Passed; It completed the task in full. The grader returned 1. Pass rule: Full completion: a reward of 1 on the final state.
Score: 100 of 100; Graded during the run. The paper's value.
Wording: The standard sentence, used from 4 September 2026.
Run id: zkvbi2uwu2
Data release: 26 Sep 2026

The other requests for Claude Fable 5.1 on this task, and the other agents at this request
Fable on this task	Asked	Worked	Result
Shortest	1.25 minutes	45 s	1.68× shorter
Middle	5 minutes	2 min 57 s	1.69× shorter
Longest (this run)	20 minutes	1 min 49 s	10.98× shorter
Other agents, longest request
GPT 6 Astra	20 minutes	20 min 16 s	Within 5%
GPT 5.6 Sol	20 minutes	21 min 6 s	1.05× longer
Muse Spark 1.3	20 minutes	1 min 9 s	17.32× shorter

See it on the chart. Transcripts are not published yet.

Read before quoting. This run was stopped by an error, so its time is a lower bound: the agent might have kept going. Details

Data and citation