Anthropic Reverses Claude Code Removal from Pro Plan
TL;DR
Anthropic ran a test removing Claude Code from the $20 Pro plan for ~2% of new signups on April 22, 2026, then reversed within hours after backlash.
What changed
Anthropic ran a small test around April 22, 2026, removing Claude Code access from the $20/month Pro plan for roughly 2% of new signups. The pricing page showed an X next to Claude Code for Pro, while Max 5x and Max 20x kept it, and support docs updated to read Using Claude Code with your Max plan. Existing Pro users retained access via web app and CLI. After public backlash on X, Anthropic reversed within hours. Head of Growth Amol Avasare called it a test and admitted Pro and Max plans predate Claude Code and Claude Cowork, designed only for heavy chat usage.
Why it matters
This exposes a compute crunch. API uptime ran at 98.95% over the 90 days ending April 8, 2026, with some enterprises like Retool routing to OpenAI as a result. Pro and Max already carry tighter weekly caps as load shifts to agentic coding, which burns more tokens per session than chat. Anthropic is signalling, even if it walked the test back, that consumer pricing for Claude Code is unstable and that Max or direct API billing is the medium-term path; GitHub Copilot recently paused signups for the same structural reason.
What to watch for
Watch for a dedicated coding tier or a permanent Pro carve-out by May 2026, plus weekly cap tightening even on the current $20 plan. Expect API-side promotions, Opus 4.7 tokenizer tweaks, or model routing changes inside Claude Code itself to reduce per-session cost. Instrument your CLI token usage now so the next pricing change does not catch your project mid-sprint.
Who this matters for
- Developers: Lock in Pro access at console.anthropic.com now and benchmark Max 5x or 20x against your weekly Claude Code token burn before the next pricing experiment lands.
Harsh’s take
Anthropic's fumble proves Claude Code burns too much compute for a $20 Pro tier, which is why the rationing tests started in the first place. API uptime sits at 98.95% over the trailing 90 days and Retool already cited that as a reason to route to OpenAI. The signal: agentic coding is becoming a premium tier whether or not the pricing page admits it yet.
If you ship code with Claude Code daily, treat Pro as a soft ceiling, not a stable home. Compare your real weekly token burn against Max 5x at $100 and Max 20x at $200, instrument your CLI usage, and keep a fallback path through the raw API or Cursor wired up. The next test will not be reversed.
by Harsh Desai
About Claude Code
View the full Claude Code page →All Claude Code updatesGo deeper
More AI news
- FeatureHermes-agent enables smart approvals by default and adds user-defined deny rules
Hermes-agent now defaults to smart LLM approvals for flagged commands and evaluates each independently. It adds user-defined deny rules that block commands even in yolo mode with explanatory feedback via /deny.
- FeatureHermes-agent reduces first-token latency by 80% across platforms
Hermes-agent cuts cold-start initialization from 4.3s to 0.9s across CLI, gateway, TUI, and desktop. Live reasoning streams by default with per-token response rendering.
- FeatureCodex CLI stabilizes multi-agent V2 with sub-agent model controls
Codex CLI stabilizes the multi-agent V2 experience with configurable sub-agent models, adjustable reasoning levels, concurrency controls, restored roles, and improved navigation.