Task 16 of 18 in ProgramBench
dandavison/delta
Sol 6.1's two runs followed the time asked: 30 min 35 s when asked for 30 min, and 8 h 1 min when asked for 8 h. Muse Wonder worked about the same whatever it was asked: 15 min 27 s when asked for 4 h, and 18 min 38 s when asked for 64 h.
- Asked for
- 22 min 30 s to 180 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 | 2 h 50 min | 53 h 41 min | 11 min 3.86× shorter asked 42 min 30 s | 10 h 24 min 1.09× shorter asked 11 h 20 min |
| Argon | 1 h 30 min | 3 h 15 min | 1 h 9 min 3.07× longer asked 22 min 30 s | 41 min 46 s 8.62× shorter asked 6 h |
| Sol 6.1 | 2 h | 1 h | 30 min 35 s Within 5% asked 30 min | 8 h 1 min Within 5% asked 8 h |
| MiMo Pro | 8 h | 1 h 31 min | 1 h 34 min 1.28× shorter asked 2 h | 1 h 8 min 28.29× shorter asked 32 h |
| MiMo UltraSpeed | 45 h | 41 min | 2 h 19 min 4.84× shorter asked 11 h 15 min | 1 h 43 min 104.83× shorter asked 180 h |
| Muse Wonder | 16 h | 9 min 20 s | 15 min 27 s 15.53× shorter asked 4 h | 18 min 38 s 206.12× shorter asked 64 h |
No runs match these filters.
Runtime is each agent’s run with no time asked.
Data and citation
Upstream: Yang et al., 2026