Anthropic's June 15 Billing Split: What Changes for Free & Paid Claude Users

If you've been running Claude inside an automation, a third-party agent tool, or via the command line, your workflow is about to cost more — or stop entirely if you're on the free tier. On May 13, 2026, Anthropic announced that starting June 15 it would divide Claude subscription usage into two separate buckets: one for typing directly in the Claude.ai chat window, and a brand-new "dedicated programmatic credit pool" for anything that calls Claude through the Agent SDK, the claude -p flag, Claude Code GitHub Actions, or third-party apps. The change is part of a broader industry pattern: The Verge reported in April that after years of subsidised access, major AI labs are under intense investor pressure to start converting usage into revenue. June 2026 is where that pressure is finally landing on ordinary users.

What Exactly Is Anthropic Changing on June 15?

Before May 13, every Claude message — whether typed manually or sent programmatically through an agent — drew from the same subscription pool. That meant a single $20 Claude Pro subscription could power both your daily chat and dozens of automated tool-use workflows. Anthropic's announcement changes that entirely. According to the official Anthropic Help Center announcement, the following usage types will be carved out into their own credit pool starting June 15:

• Agent SDK calls — any programmatic Claude usage through the official SDK
• claude -p command — piped or scripted command-line usage
• Claude Code GitHub Actions — automated code review and PR workflows
• Third-party AI agents — apps like OpenClaw, Cursor, and other tools that integrate Claude under the hood
• MCP server integrations — Model Context Protocol connections to external data sources

Direct interactive usage — opening Claude.ai and typing — is explicitly not affected. Boris Cherny, head of Claude Code, wrote on X: "Our subscriptions weren't built for the usage patterns of these third-party tools. We want to be intentional in managing our growth to continue to serve our customers sustainably long-term." The "Enter Key Test" that's circulating in developer communities is a useful rule of thumb: if a human pressed Enter to send that message, it's interactive. If code sent it, it's programmatic.

How Much Do Each Tier's Programmatic Credits Cover?

Anthropic has structured the new programmatic credit allocation to mirror the monthly subscription price. The practical monthly limits, as reported by InfoWorld, break down as follows:

The credits are billed at standard API token rates. Anthropic's own documentation notes that Claude Code averages about $6 per developer per day — meaning a Pro user's $20 monthly allowance lasts roughly three days of heavy agentic work. Once credits run dry, programmatic usage stops until the next billing cycle. Additional credits can be purchased, but at this point you're effectively paying API rates on top of your subscription fee.

For free-tier users the impact is blunt: zero programmatic credits. Tools and apps that call Claude in the background will stop working unless the app itself absorbs the API cost on the user's behalf. Some services — like Cowork, which has confirmed it will continue to be covered — are stepping in. Most won't. Check the Free Tier Tracker for ongoing updates on which integrations still pass through free Claude access after June 15.

GitHub Copilot Made a Parallel Shift on June 1 — What Happened There?

Anthropic's move didn't happen in a vacuum. Just two weeks earlier, GitHub announced that all Copilot plans transitioned to usage-based billing on June 1, replacing the previous flat-rate "premium requests" model with a token-credit system called AI Credits (1 Credit = $0.01). According to the GitHub Blog announcement, the change works like this:

• Code completions & Next Edit Suggestions — still unlimited on all plans, no credits consumed
• Copilot Chat, multi-file edits, agent tasks — now metered; each interaction burns AI Credits based on input + output + cached tokens
• Subscription prices unchanged — Pro stays $10/month, Pro+ $39/month, Business $19/user/month
• Monthly credit allowances — included with each plan, additional credits purchasable when exhausted
• Free tier — GitHub's free Copilot tier for VS Code continues, but with a smaller monthly credit allowance than paid plans

The practical effect for developers who primarily use inline completions is minimal — those stay free. But anyone who relies on Copilot Chat for lengthy code explanations, architectural reviews, or multi-file refactors will now have a visible monthly bill that scales with usage. Power users who previously hammered the chat features under a single flat subscription could see real cost increases.

Why Is This Happening Now — And Is It Just Anthropic and GitHub?

The underlying economics are not subtle. Gartner analyst Will Sommer told The Verge that capital investment in AI data centers is expected to reach $6.3 trillion between 2024 and 2029. To avoid writing down those assets, AI companies need to earn a return on invested capital of at least 7 percent — which translates to roughly $7 trillion in cumulative AI-driven revenue through 2029. That's approximately $2 trillion per year by the end of the period. For reference, the entire global software industry generates around $700 billion annually today. The math is brutal: there is no version of this story where AI stays free. The question has always been timing. OpenAI introduced in-platform advertisements. Anthropic restricted third-party tools. OpenAI also rolled out a new ChatGPT Pro tier at $100/month with 5× usage limits. Google dropped its top AI tier from $250 to $200 at I/O 2026 to stay competitive. Each of these is a different expression of the same pressure. The era of venture-subsidised AI — where the price you paid bore no relationship to the actual cost of serving your request — is ending. The companies that survive will be the ones that find the pricing sweet spot where users pay enough to fund the compute without abandoning the platform.

Which Free AI Tiers Are Still Holding Strong in Mid-2026?

Despite the squeeze, the free tier landscape isn't barren. Several providers are maintaining genuinely useful free access, either because they have different business models or because they're still in growth mode. Here's the current state as of May 31, 2026 — see the full Free Tier Tracker for live details:

• Google Gemini Flash 2.0 — free at 10 RPM via the API and unlimited in Gemini.google.com (with rate limits during peak hours)
• OpenAI GPT-4o-mini — free with daily limits in ChatGPT; API access requires a paid plan
• Meta AI (Llama-powered) — free via Meta AI apps; local Llama 3.3 70B requires your own hardware
• Mistral Le Chat — free consumer tier; developer API has a free test tier with low rate limits
• Claude.ai direct chat — still free for interactive use; programmatic access goes to zero credits June 15
• GitHub Copilot (completions) — still free and unlimited; chat features now metered
• Perplexity AI free — limited Pro searches per day; standard searches unlimited

The pattern is consistent: free tiers are surviving for interactive, rate-limited consumer chat. They are evaporating for programmatic, high-volume, or agentic use cases. If your use case is reading the news, drafting emails, or asking occasional questions, the free tier still serves you well. If you're building automations, running agents, or doing bulk processing, the era of doing that on a free subscription is essentially over. For those workflows, the honest comparison now is Free vs. Paid AI — and that comparison is increasingly settling in favour of paid.