Microsoft Research's Webwright: GPT-5.4 Success Rate Increases from 33.5% to 60.1%
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Webwright, from Microsoft Research, focuses on a model that writes code instead of chaining clicks. For long tasks, Microsoft Research reports that the same GPT-5.4 increases its success rate from 33.5% to 60.1%, producing a reusable command-line tool.
Reported Success on Long Tasks and Reusable Deliverable
Microsoft Research indicates that for long-term tasks, the GPT-5.4 model sees its success rate rise from 33.5% to 60.1% with Webwright. The system also generates a command-line tool that can be reused after the initial execution.
Principle: A Terminal to Write a Program
Webwright, developed by Microsoft Research, proposes giving the model a terminal and allowing it to write a program. This approach replaces the step-by-step interaction logic with the direct generation of executable code.
Context: The Limits of Click-by-Click
For years, web agents have operated by clicking step by step. They often failed on long tasks, and these methods produced only a sequence of actions, without a reusable deliverable.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.