You can't possibly use LLMs day to day if you think the benchmarks are solely gamed. Yes, there's been some cases, but the progress in real-life usage tracks the benchmarks overall. Gemini 2.5 Pro for example is absurdly more capable than models from a year ago.