What happened
Microsoft Research released Fara1.5, an open-weight family of AI models (4B, 9B, 27B parameters) fine-tuned on Qwen 3.5 for browser-based tasks. The Fara1.5-27B model scored 72% on the Online-Mind2Web benchmark, outperforming OpenAI Operator (58.3%) and Google's Gemini 2.5 Computer Use (57.3%). Fara1.5-9B is now live on Azure AI Foundry, with 4B and 27B variants arriving shortly, offering public weights and open inference code.
Why it matters
Access to high-performing, open-weight browser automation agents will reduce operational costs for platform engineers and procurement teams. This shifts the competitive landscape for proprietary agent providers, offering a self-hostable alternative for web scraping and form completion. Security architects gain enhanced control over agent execution environments through MagenticLite, a sandboxed browser environment that logs actions and allows user intervention. This follows OpenAI's earlier attempt with Operator, which was proprietary and later discontinued.
