GPT-5.5 Reasoning Token Clustering Issues and Benchmarking Discrepancies

Reports suggest GPT-5.5 experiences reasoning-token clustering at specific intervals, such as 516, 1034, and 1552, which may degrade performance on complex tasks. Additionally, some benchmarks indicate GPT-5.5 underperforms Claude Opus 4.8 by up to 10%.