If you let Claude Code or Codex write your Vaadin views, you'll want to know which of their models gets them right.
VaadinBench is Vaadin's benchmark for coding agents. It hands the agent a task, such as building views to a spec, bootstrapping a new project or migrating a Vaadin 14 app from Polymer to Lit. Then it runs deterministic validation tests that the agent never sees. The code passes or fails, and the model's cost in dollars is recorded. The October run made 1,440 attempts: nine tasks, five tries each, across several models and setups.
Join Vesa Nieminen, Principal Software Engineer at Vaadin, as he walks through the results.
We'll send you a confirmation email with the details.
We'll send you a confirmation email with the details.