Task 13 of 18 in ProgramBench
stranger6667/jsonschema
Sol 6.1's two runs followed the time asked: 15 min 26 s when asked for 15 min, and 4 h when asked for 4 h. Muse Wonder worked about the same whatever it was asked: 16 min 47 s when asked for 2 h 45 min, and 16 min 31 s when asked for 44 h.
- Asked for
- 15 min to 52 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 | 12 h 8 min | 33 min 31 s 1.27× shorter asked 42 min 30 s | 10 h 56 min Within 5% asked 11 h 20 min |
| Argon | 1 h 30 min | 1 h 42 min | 30 min 51 s 1.37× longer asked 22 min 30 s | 1 h 48 min 3.34× shorter asked 6 h |
| Sol 6.1 | 1 h | 34 min 39 s | 15 min 26 s Within 5% asked 15 min | 4 h Within 5% asked 4 h |
| MiMo Pro | 13 h | 1 h 47 min | 1 h 4 min 3.05× shorter asked 3 h 15 min | 1 h 56 min 26.83× shorter asked 52 h |
| MiMo UltraSpeed | 7 h | 59 min 11 s | 50 min 4 s 2.10× shorter asked 1 h 45 min | 1 h 15 min 22.35× shorter asked 28 h |
| Muse Wonder | 11 h | 8 min 50 s | 16 min 47 s 9.83× shorter asked 2 h 45 min | 16 min 31 s 159.90× shorter asked 44 h |
No runs match these filters.
Runtime is each agent’s run with no time asked.
Data and citation
Upstream: Yang et al., 2026