Claude Code briefly vanished from Anthropic's Pro plan
On April 21, 2026, Anthropic quietly removed Claude Code from the Pro plan for roughly 12 hours before rolling it back, an incident author 김이더 argues is part of a broader pattern of Anthropic struggling with the economics of long-running agentic workloads.
Score breakdown
Teams evaluating or budgeting around Claude Code for agentic workflows should watch Anthropic's plan structure and default effort settings closely, as confirmed changes to thinking budgets and a pricing experiment suggest Pro-tier access and model behavior for long-running tasks may continue to shift.
- 01On April 21, 2026, Anthropic removed Claude Code from the Pro plan, updating pricing and support docs to 'Max only', then reversed the change within ~12 hours.
- 02Anthropic's Head of Growth Amol Avasare described it as a test on ~2% of new prosumer signups, with existing Pro and Max subscribers unaffected.
- 03AMD AI director Stella Laurenzo filed GitHub issue #42796 documenting a 67% drop in Claude Code thinking depth starting in late February.
On the afternoon of April 21, 2026, Anthropic silently updated its pricing page and support documentation to remove Claude Code from the Pro plan, restricting it to Max subscribers only. The change was caught by Ed Zitron and reported by The Register. A few hours later, Anthropic rolled everything back. Anthropic's Head of Growth Amol Avasare posted on X that it was a test affecting roughly 2% of new prosumer signups and that existing Pro and Max subscribers were unaffected — but he also noted that when Max launched, it didn't include Claude Code, Cowork didn't exist, and long-running agents weren't yet common workflows. The author, 김이더, argues that framing a structural pricing reconsideration as a "small test" is telling, and that the incident is one of four connected signals that Anthropic is struggling with the economics of the agentic era.
He argued that the redact-thinking header shipped in March is a UI-only change and that Laurenzo may have been measuring a loss of external observability rather than actual reasoning degradation.
Two weeks before the Pro removal, AMD AI director Stella Laurenzo filed GitHub issue #42796 against the Claude Code repo, titled "Claude Code is unusable for complex engineering tasks with the Feb updates." Laurenzo's analysis drew on 6,852 Claude Code sessions, 234,760 tool calls, and 17,871 thinking blocks from her team over three months. Three metrics stood out: thinking depth dropped 67% on average starting in late February; the reads-per-edit ratio fell from 6.6 to 2.0 (a 70% reduction); and a stop-hook script designed to catch premature stopping and responsibility-dodging behavior fired 173 times in the 17 days after March 8, having never fired before. Laurenzo concluded that her team had moved to another provider, noting that competitors now need to be "very seriously considered and evaluated."
Claude Code team member Boris Cherny responded in the issue, offering both pushback and concession. He argued that the redact-thinking header shipped in March is a UI-only change and that Laurenzo may have been measuring a loss of external observability rather than actual reasoning degradation. However, he confirmed two substantive changes: on February 9, Opus 4.6 launched with adaptive thinking (replacing a fixed thinking budget with a per-turn model decision), and on March 3, the default effort level dropped from High to Medium (85 out of 100). The post frames all four incidents — adaptive thinking, the effort level drop, Laurenzo's data, and the Pro removal experiment — as Anthropic simultaneously testing multiple levers to manage the cost burden of long-running agentic workflows.
Key facts
- 01On April 21, 2026, Anthropic removed Claude Code from the Pro plan, updating pricing and support docs to 'Max only', then reversed the change within ~12 hours.