Agents' Last Exam, longest of three requests
GPT 5.6 Sol on “Repair a Moodle backup and rebuild registrar exports”
Shorter
10.17× shorterStopped at 10% of the time asked, 1 h 12 min short.
- Asked for
- 80 minutes
- Worked
- 7 min 52 s
- Grade
- Did not pass
- Score 95 of 100. Graded afterwards.
- How it ended
- Ended on its own
Every detail of this run
- Agent
- GPT 5.6 Sol, in Codex
- Benchmark
- Agents' Last Exam
- 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
- 7 min 52 s (472 s)
- Ratio
- 0.10× the request
- How it ended
- Ended on its own
- Grade
- Did not pass
- It did not pass the task's checker in full. The grader returned 0.95. Pass rule: Full pass: a checker score of 1.
- Score
- 95 of 100
- Graded afterwards. The paper's value.
- Wording
- The standard sentence, used from 4 September 2026.
- Run id
- vucgz6rvc5
- Data release
- 26 Sep 2026
Related runs
| Sol on this task | Asked | Worked | Split | Result |
|---|---|---|---|---|
| Shortest | 5 minutes | 9 min 36 s | 1.92× longer | |
| Middle | 20 minutes | 21 min 23 s | 1.07× longer | |
| Longest (this run) | 80 minutes | 7 min 52 s | 10.17× shorter | |
| Other agents, longest request | ||||
| GPT 6 Astra | 80 minutes | 1 h 21 min | Within 5% | |
| Claude Fable 5.1 | 80 minutes | 17 min 58 s | 4.45× shorter | |
See it on the chart. Transcripts are not published yet.
Data and citation
Upstream: Sun et al., 2026