tl;dr: Analysis of 390K Codex token_count records shows GPT-5.5 responses disproportionately terminate at exactly 516 reasoning tokens (with echoes at 1034 and 1552), accounting for 82% of such events despite only 19% of overall traffic. This clustering spiked from 0.11% in Feb 2026 to 53% in May, coinciding with a drop in mean reasoning-token usage—suggesting a hidden reasoning-budget cap, truncation, or routing behavior specific to GPT-5.5 that may explain degraded performance on complex tasks.