Task 10 of 12 in Agents' Last Exam
Apple FY2024 Balance Sheet Reconstruction
Sol 6.1's two runs came closest to the time asked: 59 s when asked for 38 s, and 10 min 23 s when asked for 10 min. Muse Wonder worked about the same whatever it was asked: 43 s when asked for 2 min, and 1 min 6 s when asked for 32 min.
- Asked for
- 30 s to 48 min
- 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 | 6 min | 4 min 35 s | 2 min 2 s 1.36× longer asked 1 min 30 s | 9 min 8 s 2.63× shorter asked 24 min |
| Argon | 2 min | 3 min 50 s | 2 min 4 s 4.12× longer asked 30 s | 3 min 6 s 2.58× shorter asked 8 min |
| Sol 6.1 | 2 min 30 s | 2 min 28 s | 59 s 1.58× longer asked 38 s | 10 min 23 s Within 5% asked 10 min |
| MiMo Pro | 12 min | 1 min 37 s | 1 min 33 s 1.94× shorter asked 3 min | 4 min 43 s 10.19× shorter asked 48 min |
| MiMo UltraSpeed | 10 min | 58 s | 50 s 2.98× shorter asked 2 min 30 s | 2 min 30 s 16.00× shorter asked 40 min |
| Muse Wonder | 8 min | 37 s | 43 s 2.78× shorter asked 2 min | 1 min 6 s 29.09× shorter asked 32 min |
No runs match these filters.
Runtime is each agent’s run with no time asked.
Data and citation
Upstream: Sun et al., 2026