is there a good metric of model degredation over time?

Im a bit lazy and only use the free models different companies host and the biggest difference i see is that some models (Gemini, OpenAI) get progressively stupid in long chats. You end up having to start a new session every oncr in a while. Or they get really hung up on a theme and cant shift to a new topic.

By contrast, Ive been impressed with Qwen. I have some chats on research and code architecture that have stretched for weeks without any noteable change in quality (though occassionally it seems to "rush" to an answer)

Im just looking at all the listed benchmarks and im unsue which i should be looking at