I gave a couple of OpenClaws out to folks. I think we’ve all absolutely loved them. But they’re extremely expensive without a subscription.
How expensive is worth spelling out, because the numbers are what changed this week. On Saturday Anthropic stopped Claude subscriptions working inside OpenClaw and other third-party harnesses, pushing that use onto “a pay-as-you-go option billed separately from your subscription.” A Claude Pro plan is $20 a month, and Max is $100 to $200. Against that, the same report carries an estimate that “a single OpenClaw agent running for one day could burn $1,000 to $5,000 in API costs.” The gap explains Anthropic’s decision and the customer’s pain in the same breath.
My Claw now runs on open-source models, and falls far short. But it also ports into an environment with Claude, GPT and Gemini subscriptions — and it still builds good code there, because my instructions are transferred verbatim to the top-tier environment. The cheap model holds the conversation. The building happens where the good models live, on plans with a fixed monthly price, and the brief arrives word for word.
OpenClaw was built to be split like this. Its README describes it as an open-source AI assistant that “runs on your own computer and meets you in the channels you already use”, says it “works with hosted and local model providers,” and adds that models and agent harnesses “are plugins you can swap without changing anything else.”
I’ll show off some OpenClaw, but the goal is to get you running on Claude Code, because that’s where you should be doing all your work from now on. Claude Code is Anthropic’s own harness — its docs describe it as “an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools.” It runs on the subscription, which is the entire point this week.
Working on a full fix. I’ll share it.
