Task 3 of 4 in PaperBench
BAM
Astra's two runs followed the time asked: 5 min 8 s when asked for 5 min, and 1 h 21 min when asked for 1 h 20 min. MiMo Pro worked about the same whatever it was asked: 2 h 38 min when asked for 2 h 44 min, and 4 h 48 min when asked for 43 h 54 min.
- Benchmark
- PaperBench
- Asked for
- 5 min to 43 h 54 min
- Runs here
- 16
- 8 agents
Every agent on this task
- Astra
- Sol
- Fable
- Opus 5.5
- Argon
- Sol 6.1
- MiMo Pro
- MiMo UltraSpeed
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 | 8 h | 13 min 40 s | 5 min 8 s Within 5% asked 5 min | 1 h 21 min Within 5% asked 1 h 20 min |
| Sol | 8 h | 8 min 44 s | 8 min 39 s 1.73× longer asked 5 min | 1 h 31 min 1.14× longer asked 1 h 20 min |
| Fable | 10 h | 11 min 34 s | 6 min 14 s 1.25× longer asked 5 min | 22 min 1 s 3.63× shorter asked 1 h 20 min |
| Opus 5.5 | 5 h | 5 h 2 min | 41 min 11 s 1.84× shorter asked 1 h 16 min | 5 h 9 min 3.92× shorter asked 20 h 12 min |
| Argon | 3 h 40 min | 1 h 33 min | 47 min 6 s 2.04× longer asked 23 min 6 s | 2 h 26 min 2.54× shorter asked 6 h 10 min |
| Sol 6.1 | 8 h | 48 min 34 s | 12 min 19 s Within 5% asked 12 min 6 s | 3 h 23 min Within 5% asked 3 h 14 min |
| MiMo Pro | 48 h | 10 h 58 min | 2 h 38 min Within 5% asked 2 h 44 min | 4 h 48 min 9.15× shorter asked 43 h 54 min |
| MiMo UltraSpeed | 48 h | 1 h 32 min | 45 min 35 s 1.97× longer asked 23 min 6 s | 3 h 41 min 1.67× shorter asked 6 h 10 min |
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: Starace et al., 2025