
Today, Pinetree Research announces a new milestone: Pinetree Agent achieves 90% on Online-Mind2Web, establishing leading performance on one of the most rigorous benchmarks for autonomous web generalization.
Online-Mind2Web evaluates whether agents can understand user intent, interpret changing interfaces, and complete real-world tasks across diverse live websites. Unlike narrow scripted benchmarks, it measures adaptability under dynamic operating conditions.
At 90%, Pinetree Agent substantially outperforms frontier systems including OpenAI Lux (83.6%), Yutori Navigator (78.7%), and Gemini 2.5 CUA (69%). This margin indicates major gains in planning, execution reliability, and cross-site generalization.
Why Online-Mind2Web Matters
Online-Mind2Web is designed to test capabilities required for practical computer use:
Intent understanding
Adaptation to unfamiliar layouts
Multi-step task execution
Robust decision-making under uncertainty
Reliable completion across changing environments
Strong performance on this benchmark suggests that autonomous systems are rapidly progressing from narrow automation toward general digital competence.
Pinetree Agent runs on Solari, Pinetree Research's cloud browsers, sandboxes and desktops for AI agents. Run your agents on Solari.