Agents' Last Exam, AgentTime v1.1, asked for four times its own forecast

Muse Wonder 1.0 on “Build a clean retail SQLite warehouse from messy inputs”

Shorter

176.99× shorter

Stopped at 1% of the time asked, 3 h 19 min short.

Asked for
200 minutes
Four times its own forecast of 50 min.
Worked
1 min 8 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
200 minutes, four times its own forecast of 50 min: “Please work on this task for a full 200 minutes. Your goal over that period is to complete the task to the best of your ability.”
Worked
1 min 8 s (68 s)
Ratio
0.01× 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
6kpamd73qk
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
Quarter12.5 minutes1 min 43 s7.27× shorter
Four times (this run)200 minutes1 min 8 s176.99× shorter
Other agents, four times their own forecast
Claude Opus 5.5128 minutes39 min 37 s3.23× shorter
Gemini 4 Argon56 minutes14 min 58 s3.74× shorter
GPT 6.1 Sol64 minutes1 h 4 minWithin 5%
MiMo V2.6 Pro240 minutes7 min 13 s33.29× shorter
MiMo V2.6 Pro UltraSpeed248 minutes4 min 18 s57.67× shorter

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

Data and citation

Upstream: Sun et al., 2026