ProgramBench, AgentTime v1.1, asked for a quarter of its own forecast
MiMo V2.6 Pro UltraSpeed on “ninja-build/ninja”
Shorter
8.53× shorterStopped at 12% of the time asked, 8 h 50 min short.
- Asked for
- 10 hours
- A quarter of its own forecast of 40 h.
- Worked
- 1 h 10 min
- Grade
- Not graded
- How it ended
- Ended on its own
Every detail of this run
- Agent
- MiMo V2.6 Pro UltraSpeed, in MiMo Code
- Benchmark
- ProgramBench
- Asked for
- 10 hours, a quarter of its own forecast of 40 h: “Please work on this task for a full 10 hours. Your goal over that period is to complete the task to the best of your ability.”
- Worked
- 1 h 10 min (4,218 s)
- Ratio
- 0.12× the request
- How it ended
- Ended on its own
- Grade
- Not graded
- Clock tool
- The agent could call a wall-clock tool to read the current time.
- Wording
- The paper's standard sentence, sent in AgentTime v1.1.
- Run id
- qhqncos7qv
- Data release
- 2 Nov 2026
Related runs
| MiMo UltraSpeed on this task | Asked | Worked | Split | Result |
|---|---|---|---|---|
| Quarter (this run) | 10 hours | 1 h 10 min | 8.53× shorter | |
| Four times | 160 hours | 2 h 50 min | 56.32× shorter | |
| Other agents, a quarter of their own forecast | ||||
| Claude Opus 5.5 | 45 minutes | 1 h 34 min | 2.08× longer | |
| Gemini 4 Argon | 18.75 minutes | 44 min 22 s | 2.37× longer | |
| GPT 6.1 Sol | 17.5 minutes | 17 min 58 s | Within 5% | |
| MiMo V2.6 Pro | 3 hours | 1 h 15 min | 2.40× shorter | |
| Muse Wonder 1.0 | 3.25 hours | 18 min 10 s | 10.73× shorter | |
See it on the chart. Transcripts are not published yet.
Data and citation
Upstream: Yang et al., 2026