Task 4 of 4 in PostTrainBench
Post-train Qwen3-1.7B-Base for Arena-Hard-v2.0 (Writing)
Sol 6.1's two runs followed the time asked: 22 min when asked for 21 min 30 s, and 5 h 46 min when asked for 5 h 44 min. MiMo UltraSpeed worked about the same whatever it was asked: 18 min 20 s when asked for 1 min 3 s, and 29 min 9 s when asked for 16 min 48 s.
- Asked for
- 1 min 3 s to 64 h 18 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 | 4 h | 16 h 4 min | 3 h 32 min 1.14× shorter asked 4 h 1 min | 44 h 46 min 1.44× shorter asked 64 h 18 min |
| Argon | 2 h 25 min | 1 h 12 min | 22 min 12 s 1.24× longer asked 17 min 54 s | 1 h 54 min 2.51× shorter asked 4 h 47 min |
| Sol 6.1 | 5 h | 1 h 26 min | 22 min Within 5% asked 21 min 30 s | 5 h 46 min Within 5% asked 5 h 44 min |
| MiMo Pro | 10 h | 2 h 24 min | 45 min 48 s 1.27× longer asked 36 min 6 s | 9 h 8 min Within 5% asked 9 h 37 min |
| MiMo UltraSpeed | 7 h | 4 min 13 s | 18 min 20 s 17.46× longer asked 1 min 3 s | 29 min 9 s 1.74× longer asked 16 min 48 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)