In light of the Opus 5.5 stealth A/B testing. What is the evidence?
retained text
Some say that the model is quantised, others say it routes Opus 5.5 to Opus 5, or Sonnet 5.5, and some others say the reasoning budgets are reduced, or double-counting tokens. What is the evidence? Also, what if we simply use Opus 5.5 Max, or Max Ultracode for everything?
document fetched 2026-11-10 00:39 · canonical https://reddit.com/r/ClaudeCode/comments/1xrlzv4/in_light_of_the_opus_55_stealth_ab_testing_what