Chinese AI Firms Copy Anthropic's Playbook on Rate Limits

Chinese AI model providers are now following Anthropic's controversial playbook of quietly cutting coding plan usage limits — and developers are not happy about it. A recent case involving a major Chinese AI company has sparked outrage after it slashed weekly API call caps by roughly 90%, just weeks after promising to honor legacy pricing.

What Happened: A Quiet Plan Downgrade

The unnamed Chinese AI provider — referred to only as 'Company G' by frustrated users — updated its user agreement on April 22 and pulled its legacy coding plan from shelves by April 30. Some users only discovered the change through a console pop-up notification on May 3, days after the old plan had already been discontinued.

The changes are dramatic:

• Old plan: 600 API calls per 5-hour window, no weekly cap
• New plan: 400 API calls per 5-hour window, 2,000 weekly cap
• Theoretical old weekly maximum: ~19,800 calls (for heavy users coding 12+ hours daily)
• New weekly maximum: 2,000 calls — a hard ceiling
• Compensation offered: 2 months of a supposedly 'equivalent' tier

The math tells a stark story. Even a moderate developer working 10 hours a day, 6 days a week, would have consumed roughly 7,200 calls weekly under the old plan. The new 2,000-call weekly cap represents a 72% reduction for moderate users and over 90% for power users.

'Designed for 5-Hour Workdays'

One developer on a popular Chinese programming forum pointed out a telling calculation: 400 calls per window × 5 hours per day = exactly 2,000 weekly calls. 'They clearly designed this product assuming developers work 5 hours a day, 5 days a week,' the user wrote.

The frustration runs deeper than the numbers. In March, the company had publicly committed to restoring legacy subscribers' right to renew their original plans. That promise lasted barely a month before the new terms replaced it entirely.

'Company A [Anthropic] acts like this, and now you do too?' the developer wrote, referencing Anthropic's well-documented history of tightening Claude usage limits for paying subscribers.

The Anthropic Precedent

The comparison to Anthropic is no coincidence. Over the past year, Anthropic has faced repeated backlash from developers over aggressive rate limiting on its Claude Pro and Max plans. Users have reported significant reductions in message allowances, often implemented without clear advance notice.

This pattern appears to be spreading across the AI industry:

• Anthropic has repeatedly adjusted Claude's usage caps downward
• OpenAI has introduced tiered rate limits across ChatGPT Plus and API access
• Chinese providers are now adopting similar strategies as costs mount
• Coding-focused plans seem to be hit hardest, likely due to high token consumption

The underlying economics are straightforward. AI coding assistants consume substantial compute resources — long context windows, frequent iterations, and multi-file analysis all drive up inference costs. As providers move past their growth-at-all-costs phase, usage restrictions become a lever for margin protection.

A Growing Trust Problem

The bigger issue may not be the rate limits themselves but how companies communicate changes. Giving users an 8-day window between a policy update and plan discontinuation — with some not even seeing the notification until after the deadline — erodes trust rapidly.

For developers who have built workflows around specific API call budgets, sudden downgrades force costly adjustments. Migration between coding AI providers is not trivial, and companies arguably exploit this switching cost when making unilateral changes.

What This Means for Developers

The trend suggests that early-adopter pricing in the AI coding tools market was never sustainable. Developers relying heavily on AI-assisted coding should consider diversifying across providers and building flexibility into their toolchains.

As the AI industry matures, expect more providers — both Western and Chinese — to follow this pattern of generous launch pricing followed by gradual tightening. The race to acquire users is giving way to the harder challenge of making unit economics work.