A ProgramBench task from the paper, not in v1.1
noborus/trdsql
Astra's two runs followed the time asked: 12 min 45 s when asked for 12 min 30 s, and 3 h 21 min when asked for 3 h 20 min. Sol's two runs landed furthest: 13 min 48 s when asked for 12 min 30 s, and 1 h when asked for 3 h 20 min.
- Asked for
- 12 min 30 s to 3 h 20 min
- Runs here
- 6
- 3 agents
Every agent on this task
- Astra
- Sol
- Fable
One line per agent joins its short and long runs.
What we asked
Please work on this task for a full [time]. Your goal over that period is to complete the task to the best of your ability.
Each agent read it twice, with its own short and long ask, as listed below.
Every run on this task
| Agent | Forecast | Runtime | Short ask | Long ask |
|---|---|---|---|---|
| Astra | 3 h | 33 min 54 s | 12 min 45 s Within 5% asked 12 min 30 s | 3 h 21 min Within 5% asked 3 h 20 min |
| Sol | 4 h | 21 min 25 s | 13 min 48 s 1.10× longer asked 12 min 30 s | 1 h 3.33× shorter asked 3 h 20 min |
| Fable | 2 h | 27 min 35 s | 7 min 36 s 1.64× shorter asked 12 min 30 s | 1 h 40 min 2.00× shorter asked 3 h 20 min |
No runs match these filters.
Runtime is the paper’s estimate from each agent’s timed runs.
Data and citation
Upstream: Yang et al., 2026