Fara1.5 browser agents are a family of open-weight browser agents developed by Microsoft Research and released in three model sizes: 4B, 9B, and 27B parameters. The family is built on Alibaba’s Qwen3.5 base model and was fine-tuned by Microsoft for browsing tasks, with all weights publicly released. On two live-web benchmarks, Online-Mind2Web and WebVoyager, Fara1.5-27B recorded scores of 72% and 88.6%, respectively.
Performance of Fara1.5 Browser Agents in Benchmark Tests
The Fara1.5 browser agents have demonstrated strong performance in benchmark tests, particularly on the Online-Mind2Web and WebVoyager platforms. The largest variant, Fara1.5-27B, scored 72% on the Online-Mind2Web test, surpassing competitors such as OpenAI Operator, which scored 58.3%, and Gemini 2.5, which achieved 57.3%. Navigator n1 and Fara1.5-9B also performed well, with scores of 64.7% and 63.4%, respectively. Additionally, on the WebVoyager benchmark, Fara1.5-27B secured a leading score of 88.6%, compared to 87.0% by OpenAI Operator and 83.0% by Holo2.
The OpenAI Operator was initially launched in January 2025 with a monthly fee of $200. However, it was eventually integrated into the ChatGPT Agent and discontinued in August. This comparative performance establishes the Fara1.5 series as a competitive choice in the field of browser agents.
Synthetic domain training was used in developing the Fara1.5 browser agents and involved creating six fully functional fake replicas of real websites to train the model on gated tasks. These six replicas were explicitly created for synthetic domain training to practice gated tasks, and they helped Fara1.5 handle ‘gated’ tasks better than predecessors.
“That’s called synthetic domain training, and it’s a significant part of why Fara1.5 handles ‘gated’ tasks better than its predecessors.”
“We started with a simple question: What does it take to make a small model genuinely good at agentic tasks? The answer spanned the full lifecycle—data generation, training objectives, model design, and orchestration had to be redesigned together rather than in isolation.”
Microsoft’s Fara1.5 browser agents combined technical innovations and targeted training to achieve strong live-web performance, outpacing OpenAI’s Operator and Google’s Gemini 2.5 on the benchmarks. The Fara1.5 family is presented as open-weight browser agents available in multiple model sizes and fine-tuned for browsing tasks. Development followed a holistic redesign of data generation, training objectives, model design, and orchestration to improve agentic task performance.
