Terminal-Bench 4.0, middle of three requests

Claude Fable 5.1 on “Repair corrupted pretraining shards and reproduce the intended checkpoint”

Shorter

1.67× shorter

Stopped at 60% of the time asked, 32 min 1 s short.

Asked

Worked

Safety cutoff 10 h, off the scale

Every detail of this run

Agent: Claude Fable 5.1, in Claude Code
Benchmark: Terminal-Bench 4.0
Task: Repair corrupted pretraining shards and reproduce the intended checkpoint
Asked for: 80 minutes: “Please work on this task for a full 80 minutes. Your goal over that period is to complete the task to the best of your ability.”
Worked: 47 min 59 s (2,879 s)
Ratio: 0.60× the request
How it ended: Ended on its own
Grade: Passed; It passed the task's tests. The grader returned 1. Pass rule: The task's tests pass.
Score: 100 of 100; Graded during the run. The paper's value.
Wording: The standard sentence, used from 4 September 2026.
Run id: fa4qewvebw
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	20 minutes	33 min 46 s	1.69× longer
Middle (this run)	80 minutes	47 min 59 s	1.67× shorter
Longest	300 minutes	1 h 12 min	4.19× shorter
Other agents, middle request
GPT 6 Astra	80 minutes	1 h 21 min	Within 5%
GPT 5.6 Sol	80 minutes	1 h 20 min	Within 5%

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

Data and citation