Task 14 of 18 in OSWorld
Blender logo and Impress
Sol 6.1's two runs followed the time asked: 10 min 17 s when asked for 9 min 30 s, and 2 h 34 min when asked for 2 h 32 min. MiMo UltraSpeed worked about the same whatever it was asked: 3 h 20 min when asked for 10 min, and 4 h 46 min when asked for 2 h 40 min.
- Asked for
- 6 min 15 s to 4 h 10 min
- Runs here
- 17
- 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 | 30 min | 41 min 24 s | 45 min 33 s 3.04× longer asked 15 min | 4 h 11 min Within 5% asked 4 h 10 min |
| Sol | 15 min | 35 min 26 s | 50 min 26 s 3.36× longer asked 15 min | 32 min 43 s 7.64× shorter asked 4 h 10 min |
| Fable | 30 min | 1 h 42 min | 1 h 24 min 5.63× longer asked 15 min | 1 h 35 min 2.62× shorter asked 4 h 10 min |
| Opus 5.5 | 50 min | 36 min 50 s | 12 min 35 s Within 5% asked 12 min 30 s | 1 h 24 min 2.38× shorter asked 3 h 20 min |
| Argon | 25 min | 2 min 38 s | 8 min 12 s 1.31× longer asked 6 min 15 s | 2 h 40 min 1.60× longer asked 1 h 40 min |
| Sol 6.1 | 38 min | 21 min 10 s | 10 min 17 s 1.08× longer asked 9 min 30 s | 2 h 34 min Within 5% asked 2 h 32 min |
| MiMo Pro | 45 min | 2 h 25 min | 1 h 31 min 8.13× longer asked 11 min 15 s | 2 h 25 min 1.24× shorter asked 3 h |
| MiMo UltraSpeed | 40 min | 6 h 14 min | 3 h 20 min 19.98× longer asked 10 min | 4 h 46 min 1.79× longer asked 2 h 40 min |
| Muse Wonder | None | None | 46 min 12 s 3.85× longer asked 12 min | No run |
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. 1 run was stopped by AgentTime, so its time is a lower bound. Details
Data and citation
Upstream: Yuan et al., 2026