Task 18 of 18 in ProgramBench
ariga/atlas
Sol 6.1's two runs followed the time asked: 25 min 25 s when asked for 25 min, and 6 h 45 min when asked for 6 h 40 min. Muse Wonder's two runs landed furthest: 14 min 10 s when asked for 4 h 30 min, and 2 min 10 s when asked for 72 h.
- Asked for
- 18 min 45 s to 120 h
- Runs here
- 12
- 6 agents
Every agent on this task
- 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 |
|---|---|---|---|---|
| Opus 5.5 | 3 h 30 min | 8 h 28 min | 41 min 47 s 1.26× shorter asked 52 min 30 s | 13 h 22 min Within 5% asked 14 h |
| Argon | 1 h 15 min | 57 min 23 s | 16 min 59 s 1.10× shorter asked 18 min 45 s | 1 h 18 min 3.86× shorter asked 5 h |
| Sol 6.1 | 1 h 40 min | 27 min 6 s | 25 min 25 s Within 5% asked 25 min | 6 h 45 min Within 5% asked 6 h 40 min |
| MiMo Pro | 30 h | 59 min 4 s | 1 h 11 min 6.31× shorter asked 7 h 30 min | 58 min 13 s 123.67× shorter asked 120 h |
| MiMo UltraSpeed | 20 h | 44 min 10 s | 44 min 7 s 6.80× shorter asked 5 h | 1 h 23 min 57.76× shorter asked 80 h |
| Muse Wonder | 18 h | 13 min 55 s | 14 min 10 s 19.05× shorter asked 4 h 30 min | 2 min 10 s 1990.78× shorter asked 72 h |
No runs match these filters.
Runtime is each agent’s run with no time asked.
Data and citation
Upstream: Yang et al., 2026