FocusBench (Meta)

Agents do personal tasks in domains such as housing, travel, IT and finance. The test checks whether they keep working through real-world blockers across several interactions as the environment changes.

Meta in-house eval behind the Active Focus feature, reported in the Muse Wonder model card. The card does not say how tasks are graded, so the scoring type is a guess. Part of Meta's PSI composite.

Made by Meta (Meta Superintelligence Labs) Released Oct 2026 General assistants Tasks kept private Last in a launch post Oct 2026 Last model added Oct 2026

Level: Everyday Most adults, without training
Scored: Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers: Agent task Many steps in a terminal, browser, desktop, app, game or robot.

Paper research.meta.ai

Models scored in lab cards

Launch posts list it, from 1 lab

Muse Wonder Full leads with 54.2%

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

Bars Table

1 Muse Wonder Full 54.2%
2 Muse Wonder 44.9%

#	Model	Score	Setting	Date	Source
1	Muse Wonder Full	54.2%		Oct 2026	Model card
2	Muse Wonder	44.9%		Oct 2026	Model card

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Reported by the lab in its model card

One lab put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Date	Launch post	Score	As the post writes it
12 Oct 2026	Muse Wonder	44.9	FocusBench (chart label)
12 Oct 2026	Muse Wonder Full	54.2	FocusBench (chart label)

Muse Wonder Full leads with 54.2%

Scores over time

One lab put it in a launch post in the last six months

More in Tool use