Understanding the New Codex Limit System After the April 9 Update

Is there any investigation for GPT-5.5 (in high mode for example)? I also found that just compressing context roughly consumed 8 % or so of the 5 hour window… That is really a lot, just for that task.