Microsoft Fara1.5: A New Era for Browser and Computer-Use Agents
microsoftai-agentsautomationcopilot-studioazure-ai-foundrycomputer-usebenchmarkwebwright
trending_upTrend: microsoft

Microsoft Fara1.5: A New Era for Browser and Computer-Use Agents

calendar_month May 22, 2026 update Updated: May 26, 2026

🔄 Update — 26 May 2026: Microsoft Webwright turns terminal web agents into reusable code

Microsoft Research has released Webwright, a terminal-native framework for steering web agents. Scoring 60.1% on the Odyssey benchmark, it makes agent workflows portable and scriptable via its deep integration with Playwright.

What’s new?

  • Webwright Framework: A terminal-native framework that enables efficient steering of web agents, turning complex interactions into reusable code.
  • Benchmark Success: Webwright scores 60.1% on the Odyssey benchmark, a significant improvement over the 33.5% achieved by base GPT-5.4 models.
  • Playwright Integration: It leverages Playwright to provide a script-first approach to browser automation, making agent behavior portable across environments.

Why this adds to the article

Webwright extends Microsoft’s computer-use agent narrative with a terminal-native layer. It demonstrates that Microsoft is focused on making web automation accessible and portable for developers, beyond just GUI-based assistants.

🔄 Update — 23 May 2026: Microsoft Fara1.5 Outperforms OpenAI Operator and Gemini in Web Benchmarks

Microsoft’s Fara1.5 has demonstrated superior performance in live-web benchmarks, surpassing leading competitors like OpenAI’s Operator and Google’s Gemini 2.5 Computer Use. This achievement reinforces Microsoft’s leadership in the browser agent space with its open-weight model.

What’s new?

  • Benchmark Dominance: Fara1.5 achieves better results in live-web scenarios than Operator and Gemini.
  • Open-Weight Advantage: Microsoft provides a powerful model that outperforms specialized competition despite its open-weight nature.

Why this adds to the article

This breakthrough confirms the maturity of the Fara1.5 model family described in the original article and demonstrates that Microsoft is not just integrating, but technologically leading the market in computer-use agents.

Summary

Microsoft is consolidating its strategy for browser and computer-use agents. With the release of the Fara1.5 model family by Microsoft Research and its integration into tools like Copilot Studio and Azure AI, the company is positioning itself as a leader in practical enterprise automation. In benchmarks such as Online Mind2Web, Fara1.5 outperforms notable competitors like OpenAI Operator and Gemini 2.5.

What happened?

Microsoft Research introduced the Fara1.5 model family, available in sizes of 4B, 9B, and 27B. These models are specifically trained to understand and interact with user interfaces in browsers and on desktop systems. Concurrently, documentation for Copilot Studio has been expanded to include “Computer Use” capabilities, signaling deep integration into the existing product landscape. The models are already listed in the Azure AI Model Catalog and are accessible to enterprise customers via Edge for Business.

Why it matters

The ability of AI models to not only generate text but also actively operate within software environments (“Action-Oriented AI”) is the next major step in the evolution of digital assistants. For businesses, this translates to massive efficiency gains in repetitive workflows that previously required manual intervention in browsers or desktop applications. Microsoft’s ability to outperform Google and OpenAI in benchmarks underscores the readiness of this technology for corporate use.

Evidence

  • Model Release: Microsoft Research published details on Fara1.5 (4B, 9B, 27B).
  • Benchmarks: On the “Online Mind2Web” benchmark, Fara1.5 achieved higher scores than OpenAI Operator and Gemini 2.0/2.5.
  • Product Documentation: Microsoft Learn now includes specific guides for computer use in Copilot Studio.
  • Availability: The Fara1.5-9B model is available for developers in the Azure AI Model Catalog.

Analysis

Microsoft’s approach distinguishes itself from purely experimental projects through its seamless embedding into the enterprise ecosystem. While other providers often present isolated “agents,” Microsoft links the Fara models directly to Edge for Business and Azure AI. This addresses critical factors such as security, compliance, and manageability—aspects that are crucial for productive use in companies. Scalability through various model sizes also enables deployment on both edge devices and in the cloud.

Practical Takeaways

  • Automation: Businesses can delegate complex browser-based processes (e.g., CRM maintenance, data extraction) to agents.
  • Integration: Utilizing Copilot Studio allows business departments without deep programming knowledge to configure agents.
  • Platform Choice: With availability in Azure AI, developers can integrate the models directly into their own applications.

Open Questions

  • Security: How will Microsoft prevent the misuse of computer-use agents for automated attacks or data exfiltration?
  • Licensing: What cost models will apply for broad corporate usage (e.g., per agent or per token)?
  • Compatibility: How robustly do the agents respond to dynamic changes in web interfaces outside the Microsoft ecosystem?

Sources

  1. Fara1.5 Research Article
  2. Copilot Studio Computer Use Docs
  3. Azure AI Model Catalog - Fara1.5-9B
  4. MarkTechPost Coverage
  5. Computerworld Report
  6. Microsoft’s Free AI Just Beat OpenAI and Google at Browsing the Web
  7. Microsoft’s Fara1.5 AI outperforms OpenAI and Google in web tasks
  8. Webwright Research Article
  9. Webwright GitHub Repo