Thanks to Michael Fairchild and Microsoft, we now have some accessibility benchmarks for a selected range of AI models – where they were tested with automatic accessibility tools after generating sample code. Project is called A11y LLM Eval, and it provides quite some insights, both directly and indirectly.