Task 3 of 18 in ProgramBench
mookid/diffr
Astra's two runs followed the time asked: 15 min 11 s when asked for 15 min, and 4 h 11 min when asked for 4 h 10 min. Muse Wonder worked about the same whatever it was asked: 23 min 23 s when asked for 1 h 15 min, and 17 min 49 s when asked for 20 h.
- Asked for
- 11 min 15 s to 28 h
- Runs here
- 18
- See Read before quoting below
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 | 41 min 36 s | 15 min 11 s Within 5% asked 15 min | 4 h 11 min Within 5% asked 4 h 10 min |
| Sol | 3 h | 22 min 41 s | 22 min 48 s 1.52× longer asked 15 min | 4 h 12 min Within 5% asked 4 h 10 min |
| Fable | 2 h | 1 h 37 min | 8 h 20 min 33.33× longer asked 15 min | 2 h 16 min 1.84× shorter asked 4 h 10 min |
| Opus 5.5 | 2 h | 3 h 7 min | 3 h 36 min 7.20× longer asked 30 min | 4 h 1 min 1.99× shorter asked 8 h |
| Argon | 45 min | 2 h 17 min | 1 h 51 min 9.85× longer asked 11 min 15 s | 1 h 26 min 2.08× shorter asked 3 h |
| Sol 6.1 | 1 h 5 min | 36 min 32 s | 16 min 33 s Within 5% asked 16 min 15 s | 4 h 20 min Within 5% asked 4 h 20 min |
| MiMo Pro | 4 h | 1 h 20 min | 1 h 39 min 1.66× longer asked 1 h | 1 h 52 min 8.54× shorter asked 16 h |
| MiMo UltraSpeed | 7 h | 59 min 2 s | 1 h 4 min 1.64× shorter asked 1 h 45 min | 1 h 13 min 22.87× shorter asked 28 h |
| Muse Wonder | 5 h | 16 min 59 s | 23 min 23 s 3.21× shorter asked 1 h 15 min | 17 min 49 s 67.34× shorter asked 20 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.
Read before quoting. 2 runs were stopped by AgentTime, so their time is a lower bound. Details
Data and citation
Upstream: Yang et al., 2026