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

Muse Wonder 1.0 on “Build a recruiting CSV and archive each candidate’s attachments”

Shorter

53.73× shorter

Stopped at 2% of the time asked, 1 h 11 min short.

Asked for
72 minutes
Four times its own forecast of 18 min.
Worked
1 min 20 s
Grade
Not graded
How it ended
Ended on its own
Every detail of this run
Agent
Muse Wonder 1.0, in Muse Code
Benchmark
AppWorld
Asked for
72 minutes, four times its own forecast of 18 min: “Please work on this task for a full 72 minutes. Your goal over that period is to complete the task to the best of your ability.”
Worked
1 min 20 s (80 s)
Ratio
0.02× 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
3qlpxts7ro
Data release
2 Nov 2026
The other ask for Muse Wonder 1.0 on this task, and the other agents at the same share of their own time
Muse Wonder on this taskAskedWorkedSplitResult
Quarter4.5 minutes56 s4.79× shorter
Four times (this run)72 minutes1 min 20 s53.73× shorter
Other agents, four times their own forecast
Claude Opus 5.544 minutes3 min 37 s12.15× shorter
Gemini 4 Argon600 seconds20 min2.00× longer
GPT 6.1 Sol24 minutes24 min 36 sWithin 5%
MiMo V2.6 Pro120 minutes8 min 31 s14.08× shorter
MiMo V2.6 Pro UltraSpeed40 minutes3 min 31 s11.40× shorter

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

Data and citation

Upstream: Trivedi et al., 2024 (ACL)