Microsoft's Fara1.5 AI Surpasses Big Tech Rivals in Web Tasks

Microsoft Research has made significant strides in artificial intelligence (AI) with the release of its Fara1.5 family, a set of open-source AI models designed for web tasks. The models have outperformed OpenAI's Operator and Google's Gemini 2.5 Computer Use on the Online-Mind2Web benchmark, demonstrating Microsoft's advancements in AI research and development.

The Fara1.5 family comprises three models with varying parameter sizes: 4B, 9B, and 27B. The flagship 27B variant scored an impressive 72% on the benchmark, while OpenAI's Operator managed 58.3% and Google's Gemini 2.5 Computer Use hit 57.3%. This achievement underscores the performance gap between Microsoft's open-source model and its proprietary counterparts.

The Qwen3.5 architecture is at the core of Fara1.5, providing a flexible framework for web interactions. Additionally, the human-in-the-loop design ensures that the agent pauses before executing critical actions, such as purchases or account changes, to prevent unintended consequences. This safeguard is particularly relevant in DeFi applications where irreversible transactions are common.