Quality Difference Between Opus 4.5 and API-Based Billing on Claude AI

The user compares their experience with Opus 4.5 on API-based billing versus the Max plan on Claude AI, noting that while the Max plan is significantly cheaper, the quality of responses has degraded over time.

Details

The user had been using Opus 4.5 on an API-based billing model, which they found to be expensive but of high quality. An intense session could cost $50, and over the last month, the user spent about twice as much as they would have on the Max plan. Intrigued by the significant cost savings, the user signed up for the Max plan on Claude AI. Initially, the Max plan seemed great, with the user only using 30% of their weekly limit despite heavy usage. However, as the user continued to use the Max plan extensively, they noticed a decline in the quality of responses, with more 'Sonnet-like' mistakes and 'doom loops' where the AI would repeat failed attempts. The user suspects that Claude AI may be 'rate limiting' or subtly changing the model parameters for Max plan users as they use the service more heavily, potentially in an effort to discourage heavy usage and drive users towards the more expensive API-based billing model.