Are you a subscriber to Anthropic’s Cloud Pro ($20 monthly) or Max ($100-$200 monthly) plans and use its Cloud AI models and products to power third-party AI agents like OpenClaw? If so, you are in for an unpleasant surprise.
Anthropic announced a few hours ago that starting tomorrow, Saturday, April 4, 2026, at 12pm PT/3pm ET, it will no longer be possible for those cloud customers to use their subscriptions to connect Anthropic’s cloud model to third-party agentic tools, citing the strain such use would place on Anthropic’s computing and engineering resources and the desire to reliably serve a large number of users.
"We are working hard to meet the growing demand for cloud, and our subscriptions were not designed for the usage patterns of these third-party tools," Boris Cherny, head of cloud code at Anthropic, wrote in a post on X. "Capacity is a resource we manage thoughtfully and we are putting our customers first by using our products and APIs."
The company has also reportedly sent an email to this effect to some customers. However, it is not certain whether cloud teams and enterprise customers will be similarly impacted. We’ve contacted Anthropic for further clarification and will update when we hear back.
To be clear, it will still be possible to use cloud models like Opus, Sonnet, and Haiku to power OpenGL and similar external agents, but users will now have to opt in to pay as they go. "additional uses" Use a billing system or Anthropic’s application programming interface (API), which charges a fee for each token used, rather than allowing open-ended usage up to certain limits, as the Pro and Max plans have allowed so far.
Reason for change: ‘Third party services are not optimized’
According to Anthropic, the technical reality is that its first-party tools like Cloud Code, its AI Vibe coding harness, and Cloud Cowork, its business app interfacing and control tool, are built to maximize "instant cache hit rates"-Reusing already processed text to save on calculations.
Third-party harnesses like OpenClaw often bypass these capabilities. “Third-party services are not optimized for this, so it’s really hard for us to do this permanently,” Cherney further explained on X.
He also revealed his own efforts to bridge the gap: “I put out some PRs specifically to improve the instant cache hit rate for OpenClave, which should help people using it with the cloud through the API/overedge.”
Prior to the news, Anthropic began enforcing strict cloud session limits every 5 hours of usage during business hours (5am-11am PT/8am-2pm ET), meaning the number of tokens you can send during those sessions was reduced.
This frustrated some power users who suddenly began reaching their limits much faster than before – Anthropic said this change was to help "Manage the growing demand for cloud" And will only affect 7% of users at any given time.
Rebates and credits to soften the blow
Anthropic isn’t banning third-party tools entirely, but it’s moving them to a separate ledger. New "additional uses" Bundles represent a middle ground between a flat-rate subscription and a full enterprise API account.
- Credit: To "soften the blow," Anthropic is offering existing customers a one-time credit equal to their monthly plan value, which can be redeemed till April 17.
-
Discount: Users who have already purchased "additional uses" Bundle can get up to one 30% offAn effort to retain power users who might otherwise churn.
-
capacity management: Anthropic’s official statement states that these devices put a "large stress" On the system, force prioritization "Customers are using our core products and APIs."
‘The all you can eat buffet just closed’
The reaction from the developer community has been a mixture of analytical acceptance and intense disappointment.
Growth marketer Akash Gupta observed on "The all-you-can-eat buffet just closed," Given that a single OpenClaw agent running for a day can incur $1,000 to $5,000 in API costs. “Anthropic was eating up that difference on every user who was rooted through a third-party harness,” Gupta wrote. “This is the speed of a company watching its margins evaporate in real time.”
However, OpenClaw’s creator, Peter Steinberger, who was recently hired by OpenAI, took a more skeptical view of it. "Capacity" logic. “Strange how the timing matches up,” Steinberger posted on X. “First they copy some popular features into their closed harness, then they lock down the open source.”
Indeed, Anthropic recently added some of the same capabilities that helped OpenGL catch up – such as the ability to send messages to agents through external services like Discord and Telegram – to cloud code.
Steinberger claimed that he and fellow investor Dave Morin had attempted "talk sense" in Anthropic, but were only able to delay enforcement by a week.
User @ashen_one, founder of Telaga Charity, expressed a concern shared by other small-scale builders: “If I switch both [OpenClaw instances] Any API key or additional usage you are recommending here would be too expensive to make it worth using. I’ll probably have to switch to a different model at this point.
.”I know it’s useless,” Cherney replied. “Fundamentally engineering is about tradeoffs, and one of the things we do to serve a lot of customers is to optimize the way subscriptions work to serve as many people as possible with the best mode.
Licensing and OpenAI Shadow
The timing of the action is particularly notable given the brain drain. When Steinberger joined OpenAI in February 2026, he brought "open paw" Ethos with that.
It appears that OpenAI is positioning itself as more "harness-friendly" Using this moment as a customer acquisition channel for alternative, potentially disgruntled cloud power users.
By limiting your membership limit "closed harness," Anthropic is claiming control over the UI/UX layer. This allows them to collect telemetry and manage rate caps more broadly, but it risks alienating the power-user community that created it. "agentic" The ecosystem comes first.
bottom line
Anthropic’s decision is a cold calculation of margin versus growth. As Cherney said, "Capacity is a resource which we manage thoughtfully."
In the 2026 AI scenario, the era of subsidized, unlimited computation is over for third-party automation.
For the average user on Claude.ai, the experience remains unchanged; The bell has rung for power consumers running autonomous offices.
<a href