Task 1 of 14 in DeepSWE
tomlkit-toml-table-converters
Sol 6.1's two runs followed the time asked: 8 min 17 s when asked for 8 min, and 2 h 9 min when asked for 2 h 8 min. MiMo UltraSpeed worked about the same whatever it was asked: 9 min 29 s when asked for 33 min 45 s, and 11 min 14 s when asked for 9 h.
- Benchmark
- DeepSWE
- Asked for
- 4 min 30 s to 10 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 | 40 min | 1 h 10 min | 49 min 10 s 4.92× longer asked 10 min | 2 h 22 min 1.13× shorter asked 2 h 40 min |
| Argon | 18 min | 1 h 2 min | 20 min 48 s 4.62× longer asked 4 min 30 s | 1 h 7 min 1.07× shorter asked 1 h 12 min |
| Sol 6.1 | 32 min | 17 min 46 s | 8 min 17 s Within 5% asked 8 min | 2 h 9 min Within 5% asked 2 h 8 min |
| MiMo Pro | 2 h | 21 min 54 s | 26 min 46 s 1.12× shorter asked 30 min | 35 min 55 s 13.37× shorter asked 8 h |
| MiMo UltraSpeed | 2 h 15 min | 11 min 29 s | 9 min 29 s 3.56× shorter asked 33 min 45 s | 11 min 14 s 48.04× shorter asked 9 h |
| Muse Wonder | 2 h 30 min | 11 min 20 s | 19 min 52 s 1.89× shorter asked 37 min 30 s | 18 min 52 s 31.80× shorter asked 10 h |
No runs match these filters.
Runtime is each agent’s run with no time asked.
Data and citation
Upstream: Huang et al., 2026