Anthropic Just Changed How You Pay for OpenClaw. Here's What It Actually Costs You Now.

Everyone's talking about the Anthropic pricing change like it's a technical infrastructure problem. Nobody's telling you what it actually means for your wallet.

TL;DR: As of April 4, your Claude Code subscription no longer covers OpenClaw usage. You now pay per use instead of a flat monthly fee. Your three options: pay as you go with Claude, switch to a different AI model, or run a free local model on your own hardware. None of these are complicated - but doing nothing means your OpenClaw setup stops working with Claude.

What Changed?

On April 4, Anthropic flipped a switch. If you were using a Claude Code subscription (Pro, Max, or Team) to power your OpenClaw setup, that subscription no longer covers it. OpenClaw founder Peter Steinberger said he tried to negotiate and only managed to delay the change by a week.

Here's the thing. Before April 4, your monthly Claude subscription included OpenClaw usage. You paid one price, used Claude through OpenClaw, and never thought about individual costs. That's gone.

Now, if you want to keep using Claude with OpenClaw, you need to set up pay-as-you-go billing through Anthropic's API. That means you pay for exactly what you use - every message, every task, every automation. Anthropic is offering full refunds to subscribers caught off guard by the switch.

Why This Matters to You

The short answer: your costs just became unpredictable.

A flat subscription is like a gym membership - you know what you're paying every month. Pay-as-you-go is like paying per rep. If you use OpenClaw casually (a few automations a day, some quick tasks), you might actually pay less than your old subscription. If you're running it hard - multiple agents, heavy automation, all day - your bill could spike without warning.

And here's the part nobody's talking about. This isn't just about money. When your costs become unpredictable, you start second-guessing every automation. "Should I run this task or is it going to cost me?" That hesitation defeats the entire point of having an AI agent.

What Are Your Three Options?

Option 1: Stay with Claude, pay as you go. Set up an Anthropic API key, add a spending cap (this is critical - do not skip the spending cap), and keep your current setup. Best for: people who use OpenClaw moderately and want to keep things simple.

Option 2: Switch to a different model. OpenClaw works with GPT-5.4, Gemini, Grok, and others. Swap your model configuration, and you're running on a different engine. Best for: people who want a fixed-cost alternative. But - and I'll cover this in a separate piece - switching models changes your security setup in ways most guides aren't telling you about.

Option 3: Run a local model for free. Google's Gemma 4 runs on consumer hardware under an Apache 2.0 license. No subscription, no API fees, no data leaving your machine. Best for: people who want zero ongoing cost and don't mind a one-time setup.

That's the cost picture. But here's what nobody in the creator space is covering yet: what actually changes in your security setup when you swap from Claude to a different model. That's next.