how does this compare to OpenAI nano class models in terms of bench mark scores?

obviously you can't run gpt 5 nano on a mobile device, but if we compare just the performance or things like UI navigation or Text extraction extraction... what is the gap. Is there really a utility gap?