Microsoft Research Releases Webwright: A Terminal-Native Web Agent Framework That Scores 60.1% on Odysseys, Up from Base GPT-5.4’s 33.5%

Microsoft Research released Webwright, an open-source framework for terminal-native web agents.

Instead of clicking through browsers step by step, agents write Playwright scripts, run bash commands, inspect logs, and iterate in a workspace.

Webwright also adds safeguards against premature completion and summarizes long histories to manage context.

In benchmarks, it outperformed a screenshot-based agent and reached 60.1% on Odysseys versus a 33.5% GPT-5.4 baseline.

The result suggests web automation agents may be more effective as code-writing terminal programs than as click-by-click browser drivers.