Hard to understand why LLM models do not seem to have much moat, but front ends + models (like ChatGPT or Claude Code) do Mistral is not one of the top LLM models anymore for at least a year, nor is it one of the top front end LLM apps either What's the idea here?