Microsoft’s Fara1.5 AI Outperforms OpenAI and Google in Browser Tasks
- Fara1.5-27B achieved a score of 72% on the Online-Mind2Web benchmark, surpassing OpenAI Operator’s score of 58.3% and Google’s Gemini’s score of 57.3%.
- The models are available in sizes of 4 billion, 9 billion, and 27 billion parameters, built on the Qwen3.5 model.
- Fara1.5-9B is currently live on Azure AI Foundry, with the other variants arriving soon.
- On the WebVoyager benchmark, Fara1.5-27B scored an impressive 88.6%, edging out OpenAI Operator’s score of 87%.
- Microsoft utilized GPT-5.4 from OpenAI to train Fara1.5 using synthetic domain training techniques.
Microsoft’s Fara1.5 AI model has made significant strides in browser task completion by outperforming competitors such as OpenAI and Google in key benchmarks like Online-Mind2Web and WebVoyager.
This achievement highlights Microsoft’s innovative training approach using synthetic domains and demonstrates their commitment to advancing open-source AI technologies for practical applications.(Source)
