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

Claude Opus 5.5 on “facebook/zstd”

Within 5%

Within 5%

Worked 99% of the time asked, within 5% of it.

Asked for
720 minutes
Four times its own forecast of 3 h.
Worked
11 h 49 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
720 minutes, four times its own forecast of 3 h: “Please work on this task for a full 720 minutes. Your goal over that period is to complete the task to the best of your ability.”
Worked
11 h 49 min (42,567 s)
Ratio
0.99× 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
tkuepxbeqh
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
Quarter45 minutes43 min 40 sWithin 5%
Four times (this run)720 minutes11 h 49 minWithin 5%
Other agents, four times their own forecast
Gemini 4 Argon300 minutes41 min 40 s7.20× shorter
GPT 6.1 Sol480 minutes8 h 1 minWithin 5%
MiMo V2.6 Pro160 hours2 h 11 min73.09× shorter
MiMo V2.6 Pro UltraSpeed140 hours56 min 41 s148.17× shorter
Muse Wonder 1.068 hours8 min 47 s464.16× shorter

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

Data and citation

Upstream: Yang et al., 2026