Microsoft Fara1.5 AI: Better Than OpenAI & Google on Web Tasks

Microsoft Research has released new AI models that browse the web better than those from OpenAI or Google. What's interesting is these models are open-weight, meaning developers can use and modify them. The Fara1.5 family includes three models. The largest, a 27B variant, scored 72% on a difficult web task benchmark. This compares to OpenAI’s Operator at 58.3% and Google’s Gemini 2.5 Computer Use at 57.3%. Essentially, Microsoft's model completed nearly three-quarters of real-world web tasks successfully. Even the 9B model scored 63.4%, outperforming both OpenAI and Google's proprietary systems. The models use a sandboxed browser interface and incorporate an "observe-think-act" loop with a human-in-the-loop safeguard. This means the AI pauses for user confirmation before critical actions like purchases. The 9B model is currently available on Microsoft Foundry. The bottom line is these open-weight models could significantly improve how AI handles complex web interactions, including those in decentralized finance, by offering enhanced performance and user safeguards.