Anthropic Removes Claude Code from Pro Subscription for New Users in Test

What Happened
On April 21, 2026, Anthropic removed access to Claude Code from its $20/month "Pro" subscription plan for new users. The changes appeared on various pricing pages, and Claude Code support documents were updated to exclusively refer to accessing Claude Code via "your Max Plan" instead of the previous "with your Pro or Max Plan."
Key Details from the Source
- Current Pro users appeared to still have access via the Claude web app during this period
- Anthropic Head of Growth Amol Avasare claimed this was a "small test of 2% of new prosumer signups"
- Support documents and website consistently showed Pro subscribers weren't offered Claude Code
- Anthropic reversed the changes to the website and support documentation at an unknown time
- Avasare stated: "...made small adjustments along the way (weekly caps, tighter limits at peak), but usage has changed a lot and our current plans weren't built for this."
- He added that Claude's Max plan was "designed for heavy chat usage, that's it" and was released before Claude Code and Claude Cowork
Context and Implications
This follows recent changes where Anthropic forced enterprise users to pay on a per-million-token basis rather than having rate limits. Some Pro users reported still being able to access Claude Code via the web app and Command-Line Interface during the test period. The article suggests this may indicate upcoming changes for all subscription tiers as Anthropic adjusts to changing usage patterns.
📖 Read the full source: HN AI Agents
👀 See Also

Self-Supervised Fine-Tuning on Own Mistakes Boosts Small Models to 80% on HumanEval
A developer trained Qwen 2.5 7B on its own self-generated coding pairs, reaching 112/164 HumanEval (+87 problems) with zero human-written training data. The approach transfers to Llama 3.2 3B and Qwen 3 4B.

Claude Code Subagents Don't Load Skills in Multi-Agent Systems
A developer reports that subagents in Claude Code v2.1.91 cannot access skills defined in .claude/skills/ directory, despite skills working perfectly in the main session. Multiple approaches including skills in agent frontmatter, Skill tool, CLI flags, and Agent Teams all fail.

Designing a Team of Agents: How Google Antigravity Structures Subagents for Autonomous Code Generation
Google Antigravity reveals its subagent architecture for autonomous coding: seven specialized agent types from the Sentinel (front-desk) to the Auditor (authenticity checker). Relevant for OpenClaw's subagent design.

Constraint Decay: Why LLM Agents Fail at Structured Backend Code
New research introduces 'constraint decay': as structural requirements accumulate, LLM agent performance drops drastically — capable agents lose 30 points in assertion pass rates, weaker ones approach zero. Actionable insights for anyone using AI coding agents.