Artificial Analysis evaluation finds Grok 4.5 matches GPT-5.5 on coding agent benchmarks at a fraction of the cost · Digg