Detailed Analysis
Anthropic conducted a limited test around April 21, 2026, temporarily restricting Claude Code access for a small portion of new Pro subscribers — the company's $20-per-month tier — by removing references to Pro plan eligibility from pricing pages and support documentation. The experiment, confirmed by Anthropic's Amol Avasare as affecting approximately 2% of new "prosumer" signups, briefly altered support documents to indicate that Claude Code required the Max plan rather than the Pro or Max plan. Crucially, existing Pro and Max subscribers were unaffected and retained full Claude Code access through the web app, desktop client, and terminal using subscription credentials rather than API keys — an important distinction that prevents users from incurring additional per-token charges beyond their subscription cost.
This test sits within a broader pattern of Anthropic incrementally tightening access controls around Claude Code as the agentic coding tool grows in adoption and resource consumption. As early as July 2025, Anthropic introduced structured weekly compute-hour rate limits targeting heavy users on both the Pro and Max tiers, effective August 28, 2025. These limits — ranging from 40–80 hours of Sonnet 4 weekly for Pro users to 240–480 hours for Max $200 subscribers — were designed to curb specific abuse vectors such as account sharing, resale of access, and the running of continuous autonomous scripts. The limits are calculated based on active compute hours rather than idle time, reflecting a token-consumption model intended to be more equitable for typical individual developers while curtailing outlier usage patterns.
The tiered access structure Anthropic is testing and refining reflects a deliberate product segmentation strategy. Pro is positioned for individual developers working with smaller codebases and standard workflows, while Max tiers ($100 and $200 per month) unlock significantly greater compute headroom and exclusive access to more capable models like Opus 4. By potentially restricting Claude Code to Max plan subscribers for new signups, Anthropic appears to be probing whether the feature can serve as an upsell mechanism — gauging price sensitivity and conversion rates among prospective heavy users without alienating its existing subscriber base. The caution with which the test was run, affecting only a narrow slice of new signups, suggests Anthropic is balancing revenue optimization against the reputational risk of restricting a flagship developer feature.
The episode also illuminates the growing tension between consumer-friendly flat-rate pricing and the economics of inference-heavy agentic AI tools. Claude Code, unlike a typical chatbot session, can execute long-running autonomous tasks that consume substantial compute over extended periods, making it fundamentally different in cost profile from standard conversational use. Community discussions have surfaced workarounds using third-party CLI tools like OpenCode, and user preferences for the official Claude Code CLI specifically because it enforces subscription token pools rather than triggering API pay-per-use billing — indicating that the pricing boundary between subscription and API access is already a friction point for sophisticated users. As agentic coding tools become central to AI-assisted software development, Anthropic's access and pricing experiments on Claude Code will likely serve as a template for how the broader industry structures consumption-based limits within flat-rate subscription products.
Read original article →