Task 4 of 18 in ProgramBench
ninja-build/ninja
Sol 6.1's two runs followed the time asked: 17 min 58 s when asked for 17 min 30 s, and 4 h 40 min when asked for 4 h 40 min. Muse Wonder worked about the same whatever it was asked: 18 min 10 s when asked for 3 h 15 min, and 12 min 4 s when asked for 52 h.
- Asked for
- 10 min to 160 h
- Runs here
- 18
- 9 agents
Every agent on this task
- Astra
- Sol
- Fable
- Opus 5.5
- Argon
- Sol 6.1
- MiMo Pro
- MiMo UltraSpeed
- Muse Wonder
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 | 2 h | 29 min 13 s | 10 min 18 s Within 5% asked 10 min | 3 h 21 min Within 5% asked 3 h 20 min |
| Sol | 4 h | 26 min 57 s | 23 min 54 s 2.39× longer asked 10 min | 37 min 32 s 5.33× shorter asked 3 h 20 min |
| Fable | 2 h 30 min | 1 h 5 min | 1 h 53 min 11.30× longer asked 10 min | 1 h 38 min 2.05× shorter asked 3 h 20 min |
| Opus 5.5 | 3 h | 4 h 23 min | 1 h 34 min 2.08× longer asked 45 min | 11 h 15 min 1.07× shorter asked 12 h |
| Argon | 1 h 15 min | 2 h | 44 min 22 s 2.37× longer asked 18 min 45 s | 1 h 19 min 3.81× shorter asked 5 h |
| Sol 6.1 | 1 h 10 min | 29 min 29 s | 17 min 58 s Within 5% asked 17 min 30 s | 4 h 40 min Within 5% asked 4 h 40 min |
| MiMo Pro | 12 h | 1 h 39 min | 1 h 15 min 2.40× shorter asked 3 h | 1 h 59 min 24.17× shorter asked 48 h |
| MiMo UltraSpeed | 40 h | 55 min 55 s | 1 h 10 min 8.53× shorter asked 10 h | 2 h 50 min 56.32× shorter asked 160 h |
| Muse Wonder | 13 h | 10 min 32 s | 18 min 10 s 10.73× shorter asked 3 h 15 min | 12 min 4 s 258.49× shorter asked 52 h |
No runs match these filters.
Runtime is a v1.1 agent’s run with no time asked, and for the paper’s three agents the paper’s estimate from their timed runs.
Data and citation
Upstream: Yang et al., 2026