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

Muse Wonder 1.0 on “AssistantBench validation task 3”

Shorter

2.78× shorter

Stopped at 36% of the time asked, 51 min 11 s short.

Asked for
80 minutes
Four times its own forecast of 20 min.
Worked
28 min 49 s
Grade
Not graded
How it ended
Ended on its own
Every detail of this run
Agent
Muse Wonder 1.0, in Muse Code
Asked for
80 minutes, four times its own forecast of 20 min: “Please work on this task for a full 80 minutes. Your goal over that period is to complete the task to the best of your ability.”
Worked
28 min 49 s (1,729 s)
Ratio
0.36× 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
jqrfj756w7
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
Quarter5 minutes5 min 43 s1.14× longer
Four times (this run)80 minutes28 min 49 s2.78× shorter
Other agents, four times their own forecast
Claude Opus 5.536 minutes36 min 29 sWithin 5%
Gemini 4 Argon840 seconds14 min 50 s1.06× longer
GPT 6.1 Sol60 minutes1 h 1 minWithin 5%
MiMo V2.6 Pro60 minutes13 min 15 s4.53× shorter
MiMo V2.6 Pro UltraSpeed120 minutes13 min 33 s8.85× shorter

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

Data and citation

Upstream: Yoran et al., 2024 (EMNLP)