“`html
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
Microsoft Research’s AI Frontiers lab has released Fara1.5, a family of computer-use agent (CUA) models for the browser. The release includes three sizes: Fara1.5-4B, Fara1.5-9B, and Fara1.5-27B. These models are integrated with MagenticLite, Microsoft’s sandboxed browser interface for these agents.
Architecture and agent loop
The models use Qwen3.5 base checkpoints in their 4B, 9B, and 27B variants. They operate through an observe-think-act loop. At each step, the model takes the prior conversation history and the three most recent browser screenshots to emit thoughts and a single next action.
The action space includes standard mouse and keyboard inputs as well as web-specific actions like web search. It also exposes meta-actions for context management such as memorizing facts for later use and asking the user clarification questions. These meta-actions allow the agent to operate over longer horizons and work collaboratively with users.
Training mix
The models are trained using supervised fine-tuning on approximately two million samples. The training data mix is 60% web trajectories, 12.8% synthetic environments, 12.5% form filling/user interactions, 8.8% grounding, and 4.9% VQA. Smaller slices cover GUI drag, instruction following, and safety. Loss is applied only to the three most recent turns in each trajectory.
FaraGen1.5: the synthetic data pipeline
FaraGen1.5 is the synthetic pipeline that produced the training trajectories. It consists of three modular components: environments, solvers, and verifiers.
Environments are split into two types: open-internet tasks run on live websites without requiring logins, while gated-domain tasks require authenticated sessions or irreversible actions like sending an email.
To handle gated domains, the team built six synthetic clones called FaraEnvs. Each clone has a realistic frontend and fully functional API with persona-based seed data in a database. These environments were created using GitHub Copilot CLI, iteratively refined by humans, ensuring that every task has a known correct outcome.
The solver agent uses OpenAI’s GPT-5.4 with custom tools mirroring the action space of Fara1.5. The solver scores 83% on Online-Mind2Web using automated WebJudge. The previous Fara-7B solver scored 67% on the same evaluation.
A user simulator is invoked when the solver issues an ask_user call or finishes a task. Three verifiers gate which trajectories enter training: correctness uses LLM-generated rubrics for open-internet tasks and privileged database judging for synthetic ones, efficiency penalizes redundant actions, and user-interaction verification checks critical points.
Critical points and safety
Fara1.5 is trained to stop and ask the user in three situations: when a task requires personal information not provided by the user; when the task description is ambiguous or lacks necessary details; and for irreversible actions without prior approval.
Safety training uses public safety datasets and internal tasks aligned with Microsoft’s Responsible AI Policy. Inside MagenticLite, all agent actions are logged and auditable. The sandboxed browser acts as a security boundary between the agent and the user’s machine.
Other benchmarks
Fara1.5-27B scores 72% on Online-Mind2Web, beating OpenAI Operator (58.3%), Gemini 2.5 CU (57.3%), and Yutori Navigator n1 (64.7%). On WebVoyager, Fara1.5-27B scores 88.6%, the 9B reaches 86.6%, and the 4B hits 80.8%. The 9B also tops similar-sized peers like MolmoWeb 8B, GUI-Owl-1.5 8B, and Holo2 8B.
On WebTailBench v1.5, Fara1.5-9B scores 64.5% process success and 32.3% outcome success. GPT-5.4 scores 79.6% process and 57.4% outcome on the same benchmark.
Key Takeaways
- Microsoft Research released Fara1.5, a family of browser computer-use agents in 4B, 9B, and 27B sizes built on Qwen3.5.
- Fara1.5-27B scores 72% on Online-Mind2Web, beating OpenAI Operator (58.3%), Gemini 2.5 CU (57.3%), and Yutori Navigator n1 (64.7%).
- The FaraGen1.5 synthetic data pipeline unlocks training on gated domains via six functional app clones (FaraEnvs) built with GitHub Copilot CLI.
- Fara1.5 pauses to ask the user at critical points: missing info, ambiguous tasks, or irreversible actions without approval.
For more details, check out the Technical details. Follow us on Twitter and join our ML SubReddit. Need to partner with us? Connect with us at this form.
“`
Originally published at marktechpost.com. Curated by AI Maestro.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
