According to CoinMeta and OneMillion.AI, Opus5.5 scored 58 points in the artificial intelligence performance evaluation, ranking first. GPT-6's Astra and Fable scored 53 points each, while Opus scored 51 points. Opus5.5 achieved the highest scores in 6 out of 10 evaluations, with an average output of about 119,000 tokens, which is about 60% more than Opus's 73,000. By reducing prices by 20% for API and by 60% for cached reads, the cost per task is 5.98 dollars, which is basically on par with Opus's 5.86 dollars. The more practical Medium category scored 51 points, with a cost of only 1.34 dollars per task. In contrast, GPT-6's SOL and Luna have performance close to their predecessors, but the prices have significantly decreased; the cost per task for SOL has dropped from 1.99 dollars to 1.06 dollars, and for Luna it has dropped from 0.18 dollars to 0.07 dollars. A divergence in their approaches is beginning to emerge: Opus5.5 focuses on enhancing its capabilities through more reasoning, whereas GPT-6 primarily aims to reduce costs.