Task 3 of 4 in PostTrainBench
Post-train Qwen3-1.7B-Base for American Invitational Mathematics Examination (AIME) 2025
Sol 6.1's two runs followed the time asked: 57 min 7 s when asked for 56 min 24 s, and 15 h 3 min when asked for 15 h 2 min. MiMo UltraSpeed's two runs landed furthest: 4 min 45 s when asked for 21 s, and 48 s when asked for 5 min 42 s.
- Asked for
- 21 s to 77 h 54 min
- Runs here
- 10
- 5 agents
Every agent on this task
- 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 |
|---|---|---|---|---|
| Opus 5.5 | 5 h | 19 h 29 min | 4 h 33 min 1.07× shorter asked 4 h 52 min | 23 h 33 min 3.31× shorter asked 77 h 54 min |
| Argon | 2 h 5 min | 5 h 48 min | 1 h 31 min Within 5% asked 1 h 27 min | 2 h 7 min 11.00× shorter asked 23 h 12 min |
| Sol 6.1 | 8 h | 3 h 46 min | 57 min 7 s Within 5% asked 56 min 24 s | 15 h 3 min Within 5% asked 15 h 2 min |
| MiMo Pro | 10 h | 20 min 59 s | 1 h 29 min 16.98× longer asked 5 min 15 s | 1 h 26 min Within 5% asked 1 h 24 min |
| MiMo UltraSpeed | 6 h | 1 min 26 s | 4 min 45 s 13.30× longer asked 21 s | 48 s 7.13× shorter asked 5 min 42 s |
No runs match these filters.
Runtime is each agent’s run with no time asked.
Data and citation
Upstream: Rank et al., 2026 (ICML)