In short: Anthropic has blocked Claude Pro and Max subscribers from using their flat-rate plans with third-party AI agent frameworks, starting with OpenClaw. The move, which took effect on 4 April 2026, shifts the cost of running autonomous agents onto users through a pay-as-you-go billing tier. The creator of OpenClaw, who joined OpenAI in February, called the decision a betrayal of open-source developers. Thousands of users now face cost increases of up to 50 times their previous monthly outlay.
Anthropic has ended a quiet subsidy that made its Claude models the engine of choice for the open-source AI agent community. Starting on 4 April 2026, users of Claude’s Pro and Max subscription tiers can no longer pipe their plan’s usage limits through third-party frameworks such as OpenClaw. If they want to keep using those tools with Claude, they must pay separately under a new “extra usage” billing system. Anthropic says it will extend the restriction to all third-party harnesses in the coming weeks.
The announcement landed as a jolt for thousands of developers who had structured their personal AI setups around the assumption that a flat monthly subscription was enough. For many of them, it no longer is.
The economics that broke the model
The logic behind the change is straightforward even if the timing was not. Claude’s subscription plans were designed around conversational use: a human opens a chat window, types a query, and reads a response. Agentic frameworks operate on a fundamentally different model. A single OpenClaw instance running autonomously for a full day, browsing the web, managing calendars, responding to messages, executing code, can consume the equivalent of $1,000 to $5,000 in API costs, depending on the task load. Under a $200-per-month Max subscription, that is an unsustainable transfer of compute costs from the user to Anthropic.
“Anthropic’s subscriptions weren’t built for the usage patterns of these third-party tools,” said Boris Cherny, Head of Claude Code at Anthropic. “Capacity is a resource we manage thoughtfully and we are prioritising our customers using our products and API.”
The scale of the problem was significant. More than 135,000 OpenClaw instances were estimated to be running at the time of the announcement, and industry analysts had noted a price gap of more than five times between what heavy agentic users paid under flat subscriptions and what equivalent usage would cost at API rates. Anthropic’s subscription business was, in effect, quietly cross-subsidising a class of usage it had not priced for.
What OpenClaw is, and why this matters
OpenClaw is an open-source AI agent framework created by Austrian developer Peter Steinberger. Originally released in November 2025 under the name Clawdbot, it was a side project: Steinberger wanted to see what would happen if you gave a large language model persistent memory, tool access, and the ability to communicate through messaging apps such as WhatsApp and Telegram. The answer, it turned out, was that an enormous number of people wanted exactly that.
The project was renamed twice in three days in late January 2026: first to Moltbot, after Anthropic raised trademark concerns about the phonetic similarity to “Claude,” and then to OpenClaw three days later. By 2 March 2026, the repository had accumulated 247,000 GitHub stars and 47,700 forks. It had become what many observers were calling the fastest-growing GitHub project in history, reaching 100,000 stars in under 48 hours at its peak. The framework supports more than 50 integrations and works across Claude, GPT-4o, Gemini, and DeepSeek. Tencent built an enterprise platform directly on top of it, demonstrating that OpenClaw’s influence had already extended well beyond individual hobbyists.
A convenient timing problem
The restriction becomes more pointed given what happened in February. On 14 February 2026, Steinberger announced he was leaving his own project to join OpenAI. Sam Altman posted publicly that Steinberger would “drive the next generation of personal agents” at the company, and that OpenClaw would be moved to an open-source foundation with OpenAI’s continued support. Steinberger wrote in a blog post that “teaming up with OpenAI is the fastest way to bring this to everyone.”
Anthropic’s restrictions were announced and enforced within weeks of that move, a timeline that has not escaped notice. Steinberger and fellow investor Dave Morin attempted to negotiate a softer landing, approaching Anthropic directly, but by their account only managed to delay enforcement by a single week.
“First they copy some popular features into their closed harness, then they lock out open source,” Steinberger wrote in response to the ban.
Whether the timing reflects competitive calculation or coincidence, the effect is the same. The most popular open-source agent framework, now loosely affiliated with OpenAI, has been effectively priced off Claude’s subscription tier.
The cost shock for users
For developers accustomed to unlimited agentic runs under a flat plan, the new billing structure is a significant disruption. Under pay-as-you-go extra usage, per-interaction costs are estimated at $0.50 to $2.00 per task, which makes heavy agentic use expensive in ways that a fixed monthly plan obscured. Some users report facing cost increases of 10 to 50 times their previous outlay. Hobbyist developers and solo practitioners, the cohort that built OpenClaw’s early adoption, are most exposed.
Anthropic has offered two concessions to smooth the transition. Subscribers receive a one-time credit equal to their monthly plan cost, redeemable until 17 April. Users who pre-purchase extra usage bundles can receive discounts of up to 30%.
Users who want to continue running OpenClaw with Claude can do so either through those extra usage bundles or by supplying a separate Claude API key, which bypasses subscription limits but charges at full API rates: $3 per million input tokens and $15 per million output tokens for Claude Sonnet 4.6, and $15 and $75 respectively for Claude Opus 4.6.
Anthropic’s closing ecosystem
The decision fits a broader pattern. Anthropic committed $100 million to its Claude Partner Network in March 2026, formalising a web of enterprise consulting and integration relationships built around its own products. Separately, the company has launched a marketplace for Claude-powered software, allowing enterprise customers to purchase third-party applications without Anthropic taking a commission, but through channels Anthropic controls. The pattern is consistent: Anthropic wants the revenue, the data, and the governance that comes with owning the customer relationship, and it is making it incrementally less attractive to route that relationship through tools it did not build.
Claude Code, Anthropic’s own developer environment, is included in Pro and Max subscription plans and is not subject to the new restrictions. The message to developers is implicit but legible: build inside Anthropic’s ecosystem, or pay API rates to build outside it.
Anthropic’s $3 billion raise in early 2026 was accompanied by language about building “artificial super-intelligence for science” and expanding its research infrastructure. What it also reflects is the commercial pressure of running one of the most computationally intensive products in the world at scale. Compute costs do not flatten because users prefer flat subscription pricing. For an AI industry that spent 2025 racing to acquire users, 2026 is increasingly about working out who actually pays for them, and how much.
Get the TNW newsletter
Get the most important tech news in your inbox each week.