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

Claude Opus 5.5 on “jesseduffield/lazygit”

Longer

2.00× longer

Worked 100% past the request, 13 h 20 min over.

Asked for
800 minutes
Four times its own forecast of 3 h 20 min.
Worked
26 h 40 min
Grade
Not graded
How it ended
Stopped at AgentTime's time limit
Every detail of this run
Agent
Claude Opus 5.5, in Claude Code
Benchmark
ProgramBench
Asked for
800 minutes, four times its own forecast of 3 h 20 min: “Please work on this task for a full 800 minutes. Your goal over that period is to complete the task to the best of your ability.”
Worked
26 h 40 min (96,000 s)
Ratio
2.00× the request
How it ended
Stopped at AgentTime's time limit
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
ajhl3zqvuu
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
Quarter50 minutes9 min 56 s5.03× shorter
Four times (this run)800 minutes26 h 40 min2.00× longer
Other agents, four times their own forecast
Gemini 4 Argon340 minutes1 h 3 min5.36× shorter
GPT 6.1 Sol480 minutes8 h 1 minWithin 5%
MiMo V2.6 Pro160 hours2 h 6 min76.42× shorter
MiMo V2.6 Pro UltraSpeed60 hours54 min 44 s65.78× shorter
Muse Wonder 1.048 hours13 min 56 s206.60× shorter

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

Read before quoting. This run was stopped at AgentTime's time limit, so its time is a lower bound: the agent might have kept going. Details

Data and citation

Upstream: Yang et al., 2026