ProgramBench, AgentTime v1.1, asked for a quarter of its own forecast

GPT 6.1 Sol on “ninja-build/ninja”

Within 5%

Within 5%

Worked 103% of the time asked, within 5% of it.

Asked for
17.5 minutes
A quarter of its own forecast of 1 h 10 min.
Worked
17 min 58 s
Grade
Not graded
How it ended
Ended on its own
Every detail of this run
Agent
GPT 6.1 Sol, in Codex
Benchmark
ProgramBench
Asked for
17.5 minutes, a quarter of its own forecast of 1 h 10 min: “Please work on this task for a full 17.5 minutes. Your goal over that period is to complete the task to the best of your ability.”
Worked
17 min 58 s (1,078 s)
Ratio
1.03× 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
b4bcrhimcz
Data release
2 Nov 2026
The other ask for GPT 6.1 Sol on this task, and the other agents at the same share of their own time
Sol 6.1 on this taskAskedWorkedSplitResult
Quarter (this run)17.5 minutes17 min 58 sWithin 5%
Four times280 minutes4 h 40 minWithin 5%
Other agents, a quarter of their own forecast
Claude Opus 5.545 minutes1 h 34 min2.08× longer
Gemini 4 Argon18.75 minutes44 min 22 s2.37× longer
MiMo V2.6 Pro3 hours1 h 15 min2.40× shorter
MiMo V2.6 Pro UltraSpeed10 hours1 h 10 min8.53× shorter
Muse Wonder 1.03.25 hours18 min 10 s10.73× shorter

See it on the chart. Transcripts are not published yet.

Data and citation

Upstream: Yang et al., 2026