ProgramBench, AgentTime v1.1, asked for four times its own forecast

Claude Opus 5.5 on “cweill/gotests”

Shorter

1.13× shorter

Stopped at 88% of the time asked, 1 h 6 min short.

Asked for
560 minutes
Four times its own forecast of 2 h 20 min.
Worked
8 h 14 min
Grade
Not graded
How it ended
Ended on its own
Every detail of this run
Agent
Claude Opus 5.5, in Claude Code
Benchmark
ProgramBench
Asked for
560 minutes, four times its own forecast of 2 h 20 min: “Please work on this task for a full 560 minutes. Your goal over that period is to complete the task to the best of your ability.”
Worked
8 h 14 min (29,626 s)
Ratio
0.88× 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
7oy5ntuj2f
Data release
2 Nov 2026
The other ask for Claude Opus 5.5 on this task, and the other agents at the same share of their own time
Opus 5.5 on this taskAskedWorkedSplitResult
Quarter35 minutes13 min 49 s2.53× shorter
Four times (this run)560 minutes8 h 14 min1.13× shorter
Other agents, four times their own forecast
Gemini 4 Argon260 minutes2 h 1 min2.15× shorter
GPT 6.1 Sol240 minutes4 h 1 minWithin 5%
MiMo V2.6 Pro28 hours1 h 35 min17.71× shorter
MiMo V2.6 Pro UltraSpeed40 hours1 h39.91× shorter

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

Data and citation

Upstream: Yang et al., 2026