MacPaw

MacPaw Research & AI team is heading to Seoul.
MacArena — our benchmark for computer-use agents on macOS — was accepted to two workshops at ICML 2026, one of the world's leading AI/ML conferences.

Full paper → http://research.macpaw.com/publications/macarena

macOS remains a genuinely hard environment for current agents, with a leading model trailing by over 26% on macOS-native tasks.
MacArena tests this across 421 manually verified tasks, 50 applications, running on Apple's native Virtualization framework on Apple Silicon.

Find the team at the poster sessions:
⬜ FAGEN Workshop, July 10, 10:10–11:00 and 14:40–15:30
⬜ AgentWild Workshop, July 11, 11:10–12:00

And if you're part of the Ukrainian research community — MacPaw is hosting a networking event at ICML.
Register here → luma.com/hk6yh6in