WildClawBench, longest of three requests
GPT 5.6 Sol on “Solve and render a color Link-a-Pix puzzle, then identify the scene”
Shorter
4.71× shorterStopped at 21% of the time asked, 23 min 38 s short.
- Asked for
- 30 minutes
- Worked
- 6 min 22 s
- Grade
- Score 70 of 100
- Graded during the run.
- How it ended
- Ended on its own
Every detail of this run
- Agent
- GPT 5.6 Sol, in Codex
- Benchmark
- WildClawBench
- Asked for
- 30 minutes: “Please work on this task for a full 30 minutes. Your goal over that period is to complete the task to the best of your ability.”
- Worked
- 6 min 22 s (382 s)
- Ratio
- 0.21× the request
- How it ended
- Ended on its own
- Grade
- 0.7
- Task grader score, 0 to 1 (a model judge only where the task calls for one). No official pass rule.
- Score
- 70 of 100
- Graded during the run. The paper's value.
- Wording
- The standard sentence, used from 4 September 2026.
- Run id
- wnjjuaabaf
- Data release
- 26 Sep 2026
Related runs
| Sol on this task | Asked | Worked | Split | Result |
|---|---|---|---|---|
| Shortest | 2 minutes | 6 min 12 s | 3.10× longer | |
| Middle | 8 minutes | 8 min 32 s | 1.07× longer | |
| Longest (this run) | 30 minutes | 6 min 22 s | 4.71× shorter | |
| Other agents, longest request | ||||
| GPT 6 Astra | 30 minutes | 30 min 15 s | Within 5% | |
| Claude Fable 5.1 | 30 minutes | 4 min 9 s | 7.22× shorter | |
See it on the chart. Transcripts are not published yet.
Data and citation
Upstream: Ding et al., 2026