Task 1 of 8 in AssistantBench
AssistantBench validation task 3
Sol 6.1's two runs came closest to the time asked: 4 min 8 s when asked for 3 min 45 s, and 1 h 1 min when asked for 1 h. MiMo Pro worked about the same whatever it was asked: 11 min 47 s when asked for 3 min 45 s, and 13 min 15 s when asked for 1 h.
- Asked for
- 53 s to 2 h
- Runs here
- 18
- 9 agents
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 | 5 min | 4 min 4 s | 1 min 40 s 1.34× longer asked 1 min 15 s | 5 min 51 s 3.42× shorter asked 20 min |
| Sol | 5 min | 4 min 55 s | 6 min 12 s 4.96× longer asked 1 min 15 s | 18 min 38 s 1.07× shorter asked 20 min |
| Fable | 3 min | 5 min 8 s | 3 min 7 s 2.50× longer asked 1 min 15 s | 8 min 6 s 2.47× shorter asked 20 min |
| Opus 5.5 | 9 min | 4 min 11 s | 2 min 42 s 1.20× longer asked 2 min 15 s | 36 min 29 s Within 5% asked 36 min |
| Argon | 3 min 30 s | 13 min 34 s | 1 min 23 s 1.58× longer asked 53 s | 14 min 50 s 1.06× longer asked 14 min |
| Sol 6.1 | 15 min | 2 min | 4 min 8 s 1.10× longer asked 3 min 45 s | 1 h 1 min Within 5% asked 1 h |
| MiMo Pro | 15 min | 6 min 51 s | 11 min 47 s 3.14× longer asked 3 min 45 s | 13 min 15 s 4.53× shorter asked 1 h |
| MiMo UltraSpeed | 30 min | 35 s | 7 min 43 s Within 5% asked 7 min 30 s | 13 min 33 s 8.85× shorter asked 2 h |
| Muse Wonder | 20 min | 1 min 19 s | 5 min 43 s 1.14× longer asked 5 min | 28 min 49 s 2.78× shorter asked 1 h 20 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: Yoran et al., 2024 (EMNLP)