Anthropic's Claude Faces Scrutiny Amidst Pricing Flips, Billing Glitches, and Quality Concerns

Anthropic recently stirred significant controversy within the developer community after users reported the sudden removal of Claude Code access from its Pro subscription tier on the official pricing page, effectively forcing a consideration of the more expensive Max plan. This change, which also extended to official documentation, was initially dismissed by Amol Abasare, Anthropic’s Head of Growth, as a ‘small test on less than 2% of new Pro user registrations.’ However, widespread user reports suggested a much larger impact, prompting Abasare to later admit the changes were ‘understandably confusing for the 98% of people not part of the experiment’ and were subsequently reverted. The incident drew sharp criticism, including a public ‘Okay, Boomer’ from OpenAI CEO Sam Altman, whose company simultaneously capitalized on the situation by promoting stable Pro plans and free Codex. Further complicating matters, users have since observed various A/B tests on Claude’s pricing, with different users seeing divergent Pro plan costs (e.g., $17 vs. $34) and Max plan adjustments, indicating ongoing, dynamic pricing experimentation.

Beyond pricing, Anthropic faced a critical billing error where a user was reportedly charged an additional $200 because Claude Code’s system detected the term ‘hermes’ (referring to a well-known open-source AI agent project, not officially supported by Claude due to perceived resource intensity) within their Git commit history, diverting them to more expensive API usage. Anthropic acknowledged this as a ‘bug with third-party detection’ and committed to refunds and credits. Compounding these issues, Anthropic has confirmed multiple instances of quality degradation in its Claude Opus 4.6 model. These included: a default reasoning capacity reduction to ‘medium’ for latency, a bug causing forgetfulness and repetitiveness in long sessions, and a system prompt change that ‘damaged code quality,’ all of which were eventually reverted after community complaints. The original speaker speculates that underlying computational resource limitations may be driving these decisions. These events contribute to a growing narrative around the high costs of AI, with some companies, as reported on Reddit, even opting for ‘AI layoffs’ by replacing AI subscriptions with human developers due to escalating expenses, and highlight a dedicated website, cloud.fail, cataloging Anthropic’s historical missteps and controversies.