Anthropic is projected to earn $18 billion this year. Some analysts estimate they have already achieved this goal ahead of schedule. But clearly, they are not satisfied.

Early this morning, a developer named George Pu posted his discovery on X: "Anthropic (parent company of Claude) just removed Claude Code from the Pro plan. Pro users who want to use this feature now must upgrade to the Max plan, with a minimum monthly fee of $100."

This means that to maintain the same level of access, the price has increased fivefold. Without a blog post, email, or changelog, Anthropic performed a very quiet page update in an attempt to raise prices.

The tweet quickly went viral. Screenshots of the pricing page began circulating on Reddit, Hacker News, and developer group chats. The community scrutinized the screen repeatedly, confirming that under the Claude Code row, the Pro column now clearly showed a cross, while only Max 5x ($100/month) and Max 20x ($200/month) retained checkmarks. Anthropic's support documentation, which previously stated "Using Claude Code with your Pro or Max plan," was changed to "Using Claude Code with your Max plan."

Naturally, the community was very dissatisfied. As public pressure mounted, Amol Avasare, Anthropic's Head of Growth, had to respond on X. His explanation was: this was only a small-scale test affecting about 2% of newly registered Pro users, and existing Pro and Max subscribers would not be affected.

Subsequently, in the same thread, he posted a longer explanation, the gist of which was: when Max was launched a year ago, it did not include Claude Code at all, Cowork didn't exist, and long-running agents were not part of daily workflows. The original design goal for Max was heavy chat users, nothing more.

However, after the release of Opus 4, usage of Claude Code exploded. Asynchronous agents became daily tools for developers. "Engagement per subscriber is way up, and our current pricing wasn't designed for this level of usage."

He finally assured: if any final plan affects existing subscribers, they will be notified well in advance, "and the message will come from us, not screenshots on X or Reddit."

Over the Past Few Months, Continuously Tightening Compute

Avasare's explanation sounds reasonable, but when pieced together, it points to a more fundamental issue: Anthropic may be under severe compute pressure.

Every Claude Code coding session consumes far more tokens than a regular conversation, sometimes by an order of magnitude. After the release of Opus 4, when users flocked to it, when agents started running asynchronous tasks for hours, and when Cowork pulled more people into heavy-usage scenarios, the marginal cost of the $20/month price point began to spiral out of control.

From a business logic perspective, this isn't a minor annoyance of "users using too much"; it's a fundamental misalignment of the pricing model after the product's capabilities evolved rapidly. Max was originally designed for heavy chat, but now Max needs to support continuously running programming agents. The compute costs for these two things are not on the same scale.

Avasare used a phrase: "engagement per subscriber is way up." Translated into business language, it means: for the same amount of money, users are now consuming significantly more resources.

This is not the first time. Looking back over a longer timeline, Anthropic has made several "compute tightening" moves in recent months.

In August 2025, Anthropic announced the introduction of weekly usage caps for Pro and Max users, effective August 28. The official reason was that some users violated terms of service by sharing and reselling accounts to circumvent limits. The statement claimed it "is expected to affect less than 5% of subscribers." Weekly limits marked the first time a hard weekly cap was imposed on high-paying users.

From late August to early September of the same year, things became more complex. Numerous developers reported on Reddit and X that Claude's code generation quality had plummeted, with the model starting to ignore its own plans, randomly producing gibberish, and Claude Code task execution becoming unreliable. Anthropic eventually admitted this was a technical bug during a reasoning stack upgrade, causing Claude Opus 4.1 to experience sustained quality degradation for about 56.5 hours, and published a formal post-mortem report.

This bug incident soon overlapped with a larger controversy.

In early March 2026, Stella Laurenzo, a Senior AI Director at AMD, published a detailed analysis on GitHub. Based on 6,852 Claude Code session files, 17,871 thinking blocks, and over 230,000 tool calls, the conclusion was: Claude Code had gotten dumber. This analysis resonated strongly within the developer community.

Boris Cherny, Head of Claude Code, later responded, denying the core conclusion of "model degradation." He explained that Opus 4.6 switched to an "adaptive thinking" mode on February 9th, and on March 3rd, the default effort level was adjusted to medium (85). The official stance was that this represented the "best balance between intelligence, latency, and cost." However, he also admitted that about 7% of users would now hit session limits they previously wouldn't have, and this adjustment primarily targeted Pro users during peak hours, with Team and Enterprise users unaffected.

From March 13 to 27, 2026, Anthropic launched a "double usage during off-peak hours" promotion: usage limits were automatically doubled on weekdays from 2 PM to 8 AM ET the next day, and throughout the entire weekend, covering all plans: Free, Pro, Max, and Team.

On the surface, it was a benefit. But considering the context, the doubling was placed in off-peak hours precisely because peak hours were already under compute strain. Giving double capacity to idle resources doesn't affect congested periods while also creating an incentive to upgrade to higher-tier plans.

Then, on April 4th, Anthropic announced a ban on third-party agent frameworks running using Pro and Max subscription quotas. Users who wanted to continue calling Claude through external frameworks like OpenClaw would have to pay extra on a per-use basis.

And then came this incident of Claude Code disappearing from the Pro pricing page.

Anthropic Certainly Won't Be the Last to Do This

Returning to Avasare's tweet. He said "Claude Code disappearing from the Pro pricing page is a small-scale test affecting only about 2% of newly registered Pro users," but this statement sparked widespread skepticism in the community.

The most direct question came from developer Simon Willison: "Everyone I know saw the new pricing page, and the Internet Archive has already saved a screenshot." If it truly only affected 2% of users, why was the entire public pricing grid changed? Why were support documents updated site-wide?

Even more intriguing: when a reporter pressed Avasare on why the pricing page and support documentation were comprehensively altered—a scope impossible to explain with a test targeting only 2% of users—he did not respond. An Anthropic company spokesperson also did not answer this follow-up question.

Another contradiction was noted: Claude Cowork was not removed from the Pro plan, but from a product functionality perspective, Cowork is essentially Claude Code under a different brand name. As one developer put it, "Cowork is Claude Code wearing a less intimidating outfit."

The outcome was: after several hours of舆论发酵, Anthropic quietly reverted the website and support documentation to their original state. No explanation, no announcement. Just as quietly as the change was made, it was undone.

If we look at all these actions together, we find that Anthropic's intentions are now as clear as day to everyone:

Anthropic is using a probing approach, step by step testing the boundaries of higher compute usage with its subscribers. Doubling on weekends guides demand towards idle times; lowering the default effort level saves compute during peak hours; banning third-party agent frameworks plugs quota arbitrage loopholes; and removing Claude Code from the Pro page tests user reaction to more aggressive repricing strategies.

Amol himself admitted: "Our current pricing wasn't designed for this level of usage." This is a rare admission. It indicates that Anthropic has, to some extent, realized the current pricing structure cannot sustainably support the compute consumption brought by Claude Code.

What will change next, we don't know.

But perhaps, as Ed Zitron wrote in his article: "I don't think this is the last time Anthropic will do this."