benchmark
Benchmark for testing autonomous web agents on realistic tasks across simulated e-commerce, forum, and admin websites.
Microsoft Research's Webwright lifted GPT-5.4 from 33.5% to 60.1% on 200 long web tasks by having it write browsing code from a terminal. Because the baseline steered by screen coordinates, the result compares writing code with clicking pixels.
Reality
Berkeley's RDI center attacked the step where each benchmark computes its score, and without solving a task its own scorecard reports 100% on five of the eight, about 98% on GAIA and 73% on OSWorld.
Reality
The mitigation vendors cite when asked about prompt injection is instruction hierarchy. In WASP's sandbox, agents built on models that have it still began following text a human wrote into a webpage.
Reality
A dev.to write-up puts agent-browser inside a UI evaluation harness so a coding agent hunts for the Settings dialog through an accessibility snapshot, with a one-line Playwright assertion deciding pass or fail.
Reality
A new diagnostic benchmark treats the execution layer as something to vary rather than a fixed backdrop. Its conclusion is that agent capability belongs to a model-harness pair.
Reality
A preprint from ServiceNow AI Research and university co-authors reports that a plain web agent with a longer horizon matches or beats AWM, ASI and ReasoningBank, often on fewer tokens.
Reality