Looks promising. Let's hope that the benchmarks and experiments for DeepSeek are truly done independently and not tainted or paid for by them (Unlike OpenAI with FrontierMath.)