Similarly I found the results pretty mixed whenever a library or framework with a lot of releases/versions is involved. The LLM tends to mix and match features from across versions.