OSWorld 2.0, longest of three requests

GPT 5.6 Sol on “Reimbursement package”

Shorter

24.33× shorter

Stopped at 4% of the time asked, 4 h short.

Asked

Worked

Safety cutoff 8 h 20 min, off the scale

Every detail of this run

Agent: GPT 5.6 Sol, in Codex
Benchmark: OSWorld 2.0
Task: Reimbursement package
Asked for: 250 minutes: “Please work on this task for a full 250 minutes. Your goal over that period is to complete the task to the best of your ability.”
Worked: 10 min 16 s (616 s)
Ratio: 0.04× the request
How it ended: Ended on its own
Grade: Did not pass; It did not meet every checkpoint. The grader returned 0. Pass rule: Binary completion: a score of 1.00, with every checkpoint met.
Score: 0 of 100; Graded during the run. The paper's value.
Wording: The standard sentence, used from 4 September 2026.
Run id: uvbs2wd2vq
Data release: 26 Sep 2026

The other requests for GPT 5.6 Sol on this task, and the other agents at this request
Sol on this task	Asked	Worked	Result
Shortest	15 minutes	30 min 3 s	2.00× longer
Middle	60 minutes	1 h 1 min	Within 5%
Longest (this run)	250 minutes	10 min 16 s	24.33× shorter
Other agents, longest request
GPT 6 Astra	250 minutes	4 h 11 min	Within 5%
Claude Fable 5.1	250 minutes	46 min 39 s	5.36× shorter

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

Data and citation