ProgramBench, AgentTime v1.1, asked for four times its own forecast
Gemini 4 Argon on “ninja-build/ninja”
Shorter
3.81× shorterStopped at 26% of the time asked, 3 h 41 min short.
- Asked for
- 300 minutes
- Four times its own forecast of 1 h 15 min.
- Worked
- 1 h 19 min
- Grade
- Not graded
- How it ended
- Ended on its own
Every detail of this run
- Agent
- Gemini 4 Argon, in Antigravity CLI
- Benchmark
- ProgramBench
- Asked for
- 300 minutes, four times its own forecast of 1 h 15 min: “Please work on this task for a full 300 minutes. Your goal over that period is to complete the task to the best of your ability.”
- Worked
- 1 h 19 min (4,722 s)
- Ratio
- 0.26× 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
- cnfenn6sxq
- Data release
- 2 Nov 2026
Related runs
| Argon on this task | Asked | Worked | Split | Result |
|---|---|---|---|---|
| Quarter | 18.75 minutes | 44 min 22 s | 2.37× longer | |
| Four times (this run) | 300 minutes | 1 h 19 min | 3.81× shorter | |
| Other agents, four times their own forecast | ||||
| Claude Opus 5.5 | 720 minutes | 11 h 15 min | 1.07× shorter | |
| GPT 6.1 Sol | 280 minutes | 4 h 40 min | Within 5% | |
| MiMo V2.6 Pro | 48 hours | 1 h 59 min | 24.17× shorter | |
| MiMo V2.6 Pro UltraSpeed | 160 hours | 2 h 50 min | 56.32× shorter | |
| Muse Wonder 1.0 | 52 hours | 12 min 4 s | 258.49× shorter | |
See it on the chart. Transcripts are not published yet.
Data and citation
Upstream: Yang et al., 2026