Your $20/month Claude Pro plan no longer covers third-party agents. The era of flat-rate AI agents is over.
If you’ve been running OpenClaw on your Claude Pro subscription and wondering why it felt too good to be true, Anthropic just confirmed that it was.
Starting today at 12pm PT, Claude subscriptions no longer cover usage through third-party tools. That means OpenClaw, NanoClaw, OpenCode, and anything else that was piggybacking on your $20 or $200 monthly plan through OAuth authentication is now cut off from your included limits.
You can still use Claude to power these tools. You just have to pay separately now. Either through “Extra Usage” bundles (pay as you go, billed on top of your subscription) or through a standard API key with per-token pricing.
Anthropic buried the announcement on a Friday night. The enforcement starts on a Saturday. Users got less than 24 hours notice.
Why This Is Happening
The short version: OpenClaw was eating Anthropic’s lunch.
Claude subscriptions assume average usage patterns. A person opens the chat, asks a few questions, closes the tab. OpenClaw doesn’t work like that. It runs 24 hours a day, sends requests in the background through the Heartbeat scheduler, processes multiple agent workflows simultaneously, and never stops. As The Decoder put it, “OpenClaw is like a sumo wrestler at an all-you-can-eat buffet.”
Boris Cherny, Head of Claude Code at Anthropic, said their subscriptions “weren’t built for the usage patterns of these third-party tools” and that capacity management required prioritizing their own products.
Translation: Claude Code and Cowork need those servers. OpenClaw was taking too much of the pie.
The Timing Is Not Subtle
OpenClaw creator Peter Steinberger left for OpenAI in February 2026. Within weeks, Anthropic updated its terms to explicitly prohibit subscription OAuth tokens in third-party tools. Then they shipped Cowork and Claude Code Channels, both of which replicate core OpenClaw features (phone to desktop dispatch, messaging channel integration). Now they’ve cut off the billing pathway.
Steinberger called it out directly: “Funny how timings match up, first they copy some popular features into their closed harness, then they lock out open source.”
He and OpenClaw board member Dave Morin tried to negotiate with Anthropic. The best they got was a one week delay.
What It Costs Now
Here’s the math that matters.
If you were running OpenClaw on a $20/month Claude Pro plan, you were getting effectively unlimited agent usage for a flat rate. That’s over. Under the new model, every request through OpenClaw burns API tokens at standard rates.
For context, from our OpenClaw deep dive: a light user running email triage and basic automations spends $6 to $13/month on API costs. A regular user with multiple daily workflows on Claude Sonnet hits $15 to $30/month. Heavy users with premium models and browser automation can reach $50 to $200/month. One documented case hit $3,600/month from unmonitored runaway automations.
So your “$20/month AI assistant” just became a “$20 subscription plus $15 to $200 in API fees” situation. For a lot of casual OpenClaw users, the math stops working.
What You Should Do
If you’re currently running OpenClaw on Claude, you have four options:
Option 1: Enable Extra Usage. Stay on your Claude subscription, turn on pay-as-you-go billing in your account settings. Anthropic is offering up to 30% off pre-purchased bundles and a one-time credit equal to your monthly plan cost (redeemable by April 17). This is the path of least resistance.
Option 2: Switch to API key billing. Get a key from console.anthropic.com, set a daily spending limit, and use that instead of your subscription auth. Cleaner billing, more control, same Claude models. Set your default model to Sonnet, not Opus, unless you want surprise bills.
Option 3: Switch providers. OpenClaw works with every major model. OpenAI’s Codex subscriptions still support third-party harnesses (for now). OpenAI employee Thibault Sottiaux explicitly endorsed this on X the same day Anthropic made the announcement. Alibaba Cloud, MiniMax, and Z.AI also offer coding plans that work with OpenClaw. Claude might have the best models, but it’s no longer the cheapest way to power an always-on agent.
Option 4: Run local models. Ollama with a capable open model like Gemma 4 (just released under Apache 2.0) costs $0 in API fees. You need decent hardware (Mac Mini M4 with 24GB RAM is the community favorite) but the ongoing costs drop to near zero. Less capable than Claude for complex reasoning but perfectly fine for routine automation tasks.
The Bigger Pattern
This isn’t just an Anthropic story. It’s the end of the all-you-can-eat era for AI agents.
Google already restricted Gemini CLI access for third parties. Microsoft pushes Azure users toward its own AI services. OpenAI has tightened third-party rules over the past year (though they’re currently playing the hero by welcoming OpenClaw users, which is convenient positioning more than principled commitment to open source).
The economics were always going to break. Flat-rate subscriptions and 24/7 autonomous agents pulling thousands of API calls per day were never going to coexist peacefully. Automation platforms charge per execution for the same reason: unlimited plans collapse when users actually use them unlimitedly.
Every AI company that offers a flat-rate subscription is watching what just happened to Anthropic and doing the same math in their heads right now. OpenAI is welcoming OpenClaw users today. The question is whether they’ll still be welcoming them in six months when the compute bills land.
If you’re building anything on top of these tools, the lesson is clear: own your billing relationship with the model provider directly. Don’t route it through someone else’s subscription loophole. Because the loophole always closes.
