Mark Chen's evaluations show GPT-5.5 outperforms Claude Opus 4.8 by 20.7% in high-configuration tests at half the cost

GPT-5.5 also ran at double the execution speed.