Microsoft Research has announced Webwright, a terminal-native browser agent framework that utilizes GPT-5.4 to automate web tasks. The framework achieved a 60.1% score on the long-horizon Odysseys benchmark, setting a new high among open-sourced web automation harnesses.
Microsoft Research has introduced Webwright, a new terminal-native browser agent framework designed to replace traditional click-trace web automation with reusable Playwright scripts. The framework is powered by GPT-5.4 and operates using a single agent loop across three modules. Webwright achieved a 60.1% score on the long-horizon Odysseys benchmark and an 86.7% score on Online-Mind2Web, marking it as the highest AutoEval score among open-sourced harness recipes.