Last week, a move by Anthropic caused an uproar among many heavy Claude users: starting April 4, Claude Pro and Max subscribers can no longer use fixed monthly fees to run third-party agent frameworks like OpenClaw. To continue using them, they must switch to pay-as-you-go pricing.
When the news broke, my first reaction was: this makes sense, but the timing is very delicate.
What exactly is OpenClaw
If you're not in the AI tools space, you might not have heard of OpenClaw. Simply put, it's an open-source AI agent framework specifically designed to let Claude run complex automation tasks locally—writing code, calling APIs, running workflows—a combination that can replace a lot of repetitive work.
By March 2026, the project had accumulated 247,000 stars on GitHub, making it one of the fastest-growing AI tools. With more users, the traffic hitting Anthropic's servers was no longer ordinary chat requests, but heavy tasks that could run for hours and consume thousands of tokens.
The real impact of the new pricing
Anthropic didn't explicitly say how much prices would increase, but developers quickly did the math:
| Usage pattern | Before (monthly subscription) | Now (pay-as-you-go) |
|---|---|---|
| Light usage | $0-$100/month | Roughly the same |
| Moderate agent tasks | $0-$100/month | $100-$500/month |
| Running OpenClaw 24/7 | $0-$100/month | $1,000-$5,000/month |
Some developers reported on forums that their monthly costs had increased by 10 to 50 times. This is no small change.
Anthropic offered two compensation measures: a one-time credit equal to one month's subscription fee (to be used by April 17), and up to 30% discounts on prepaid usage packages. But for teams that had already integrated OpenClaw into their production environments, this buffer is essentially a drop in the bucket.
Does Anthropic's reasoning hold up?
Claude Code lead Boris Cherny said:
Our subscription plans were never designed for this kind of third-party tool usage pattern.
That statement is actually correct. The logic of a monthly subscription is: you use it dozens of times a day, Anthropic absorbs the API costs, and the average usage makes it worthwhile for everyone. But if someone uses the same monthly fee to run computing power worth thousands of dollars, the math simply doesn't add up.
The problem is, did Anthropic explicitly say this wasn't allowed before? No. Many developers only encountered this restriction after OpenClaw exploded in popularity and became their primary workflow. Changing the rules with a blanket cut, even if logically justified, is a poor execution.
The timing is quite telling
Another point of discussion: OpenClaw's original author joined OpenAI in February 2026.
Anthropic's restrictions landed just a few weeks after that.
I have no evidence that the two events are directly related. But if you're a heavy Claude user, seeing this timeline makes it hard not to wonder.
Developers' options
The reality is:
- Light users: Not much impact, can continue with subscriptions
- Heavy agent users: Either accept the new pricing or switch to other models
- Cost-saving direction: DeepSeek V4 and Qwen 3.5 cost about one-twentieth of GPT-5 for agent tasks, and the performance gap is narrowing
For Anthropic, this can control costs in the short term. But if developers migrate en masse to open-source alternatives due to pricing, the ecosystem loss will be harder to recover from.
After all, behind a framework with 247,000 stars is a very large base of paying users.
Sources: Anthropic blocks OpenClaw from Claude subscriptions in cost crackdown (TNW), CocoLoop, Anthropic says Claude Code subscribers will need to pay extra for OpenClaw usage (TechCrunch)