ProgramBench, AgentTime v1.1, asked for four times its own forecast
MiMo V2.6 Pro UltraSpeed on “lfos/calcurse”
Shorter
21.06× shorterStopped at 5% of the time asked, 34 h 17 min short.
- Asked for
- 36 hours
- Four times its own forecast of 9 h.
- Worked
- 1 h 43 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
- Task
- lfos/calcurse
- Asked for
- 36 hours, four times its own forecast of 9 h: “Please work on this task for a full 36 hours. Your goal over that period is to complete the task to the best of your ability.”
- Worked
- 1 h 43 min (6,155 s)
- Ratio
- 0.05× 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
- qsyavzu4xe
- Data release
- 2 Nov 2026
Related runs
| MiMo UltraSpeed on this task | Asked | Worked | Split | Result |
|---|---|---|---|---|
| Quarter | 2.25 hours | 1 h 21 min | 1.66× shorter | |
| Four times (this run) | 36 hours | 1 h 43 min | 21.06× shorter | |
| Other agents, four times their own forecast | ||||
| Claude Opus 5.5 | 660 minutes | 10 h 35 min | Within 5% | |
| Gemini 4 Argon | 360 minutes | 1 h 34 min | 3.82× shorter | |
| GPT 6.1 Sol | 360 minutes | 6 h | Within 5% | |
| MiMo V2.6 Pro | 60 hours | 1 h 31 min | 39.58× shorter | |
See it on the chart. Transcripts are not published yet.
Data and citation
Upstream: Yang et al., 2026