Microsoft Releases Fara1.5: A Family of Browser Computer-Use Agents (4B/9B/27B) That Outperform OpenAI Operator and Gemini 2.5 Computer Use on Online-Mind2Web

Editor's take

Microsoft Research has introduced Fara1.5, a new set of browser-based AI agents that demonstrate strong performance in simulating human interaction with web applications. The 27 billion parameter version achieved a 72% accuracy on the Online-Mind2Web benchmark, surpassing established models like OpenAI's Operator and Google's Gemini 2.5 Computer Use.

This development is significant as it showcases progress in building more capable and efficient AI agents for web navigation and task completion. Such agents could eventually streamline online workflows, automate repetitive digital tasks, and enhance accessibility for users by providing more intelligent web assistants. The performance improvement over prior benchmarks suggests a move towards more reliable and sophisticated agents for real-world web interactions.

Future developments to monitor include the agent's ability to generalize to a wider array of complex, dynamic web interfaces beyond the benchmark's scope and its practical integration into user-facing products. The efficiency and cost-effectiveness of deploying these agents at scale, particularly the larger models, will also be a key indicator of their broader impact.