ProgramBench, longest of three requests
Claude Fable 5.1 on “ninja-build/ninja”
Shorter
2.05× shorterStopped at 49% of the time asked, 1 h 42 min short.
- Asked for
- 200 minutes
- Worked
- 1 h 38 min
- Grade
- Did not pass
- Score 95.8 of 100. Graded afterwards.
- How it ended
- Ended on its own
Every detail of this run
- Agent
- Claude Fable 5.1, in Claude Code
- Benchmark
- ProgramBench
- Asked for
- 200 minutes: “Please work on this task for a full 200 minutes. Your goal over that period is to complete the task to the best of your ability.”
- Worked
- 1 h 38 min (5,855 s)
- Ratio
- 0.49× the request
- How it ended
- Ended on its own
- Grade
- Did not pass
- It did not pass every hidden test. The grader returned 0.958. Pass rule: Resolved: every hidden test passes.
- Score
- 95.8 of 100
- Graded afterwards. The paper's value.
- Wording
- The standard sentence, used from 4 September 2026.
- Run id
- rw4utnz4x2
- Data release
- 26 Sep 2026
Related runs
| Fable on this task | Asked | Worked | Split | Result |
|---|---|---|---|---|
| Shortest | 10 minutes | 1 h 53 min | 11.30× longer | |
| Middle | 40 minutes | 26 min 59 s | 1.48× shorter | |
| Longest (this run) | 200 minutes | 1 h 38 min | 2.05× shorter | |
| Other agents, longest request | ||||
| GPT 6 Astra | 200 minutes | 3 h 21 min | Within 5% | |
| GPT 5.6 Sol | 200 minutes | 37 min 32 s | 5.33× shorter | |
See it on the chart. Transcripts are not published yet.
Data and citation
Upstream: Yang et al., 2026