# Anthropic Programmatic Credits: How the New Metered System Works and What It Costs On June 15, 2024, Anthropic changed how programmatic access works for Claude subscription plans. Instead of the previous near-unlimited model, **subscribers now receive a monthly dollar-denominated credit budget specifically for non-interactive use**: the Agent SDK, the `claude -p` CLI, and approved third-party applications. Credits are consumed at standard public API rates and do not roll over. ## Background: the previous model and why it changed Anthropic's Pro and Max subscriptions previously permitted programmatic use alongside interactive chat access for a flat monthly fee. Developers could run the Agent SDK, use `claude -p` in scripts and CI/CD pipelines, and authenticate third-party tools built on Claude, all under the same subscription. Tools including OpenClaw (an inbox and calendar agent), Sandcastle (TypeScript agent orchestration), T3 Code (a multi-agent desktop control plane), and Zed (a code editor with built-in Claude integration) all ran on user subscriptions. The model was not sustainable at the high end. Some Max plan users were consuming workloads worth tens of thousands of dollars at standard API rates while paying $200 per month. Anthropic absorbed the difference. ### The crackdown On January 9, 2024, without prior announcement, Anthropic silently blocked subscription authentication tokens from working in non-official applications. Third-party tools stopped working overnight. In February 2024, Anthropic updated its Terms of Service to formally prohibit "third-party harnesses." In April 2024, enforcement began, and the Claude Code system prompt was updated to scan the user's local Git repository for keywords associated with specific tools (including "OpenClaw" and "Hermes") as a detection mechanism. ![Blog post titled "Anthropic's Walled Garden: The Claude Code Crackdown" detailing the sudden block of third-party tools](https://imagedelivery.net/xZXo0QFi-1_4Zimer-T0XQ/8f814415-848d-4bcc-9749-23e38e4fb400/lg2x =1920x1080) The Git scanning was widely criticized as a fragile proxy: developers could be flagged for having a commit message or file mentioning a tool name, regardless of whether they were actively using it. The programmatic credits system replaces this period of informal enforcement with an explicit, metered model. ## How programmatic credits work Programmatic credits are separate from the interactive usage limits on claude.ai. The message rate limits for the chat interface are unchanged. Credits apply only to: - The **Agent SDK** (Python and TypeScript) - The **`claude -p`** CLI command - Approved **third-party applications** ### Credit amounts by plan ![Monthly credit by plan table showing credit amount for each subscription tier](https://imagedelivery.net/xZXo0QFi-1_4Zimer-T0XQ/cb9aa83a-f7ff-4443-0607-c66fbfaa7000/md1x =1920x1080) | Plan | Monthly subscription | Monthly programmatic credit | |---|---|---| | Pro | $20 | $20 | | Max 5x | $100 | $100 | | Max 20x | $200 | $200 | | Team | $20/seat | $20/seat | | Team Premium | $100/seat | $100/seat | Credits refresh at the start of each billing cycle. Unused credits do not carry over. ### Billing at standard API rates ![Anthropic model pricing table showing cost per million input and output tokens for models including Claude Opus 4.7](https://imagedelivery.net/xZXo0QFi-1_4Zimer-T0XQ/235e2337-cf83-4253-cad9-d9178ae92500/orig =1920x1080) Credits are consumed at Anthropic's public API rates: per million input tokens and per million output tokens. This is identical to pay-as-you-go API billing. A task requiring many tokens on Claude Opus 4.7 costs the same whether billed from programmatic credits or a direct API account. For a Pro plan user spending the $20 credit on Claude Opus 4.7, a moderately complex coding session can exhaust the monthly budget in a single afternoon. The $20 credit is not a month of programmatic access; it is $20 of API usage billed at full API rates. ### When credits are exhausted If the monthly credit is depleted before the billing cycle ends, programmatic access stops until credits refresh. Optionally, users can enable "usage credits," which authorizes Anthropic to charge any overage against the payment method on file, converting excess use to pay-as-you-go billing on top of the subscription fee. ## Practical cost implications For the Pro plan at $20/month, the credit provides roughly the same dollar value as a direct API account with $20 loaded. For simple scripting or occasional use, this may be sufficient. For any sustained or complex agentic workload on a capable model, it is not. For the Max 20x plan at $200/month, the credit was a meaningful budget in the previous environment, where some users consumed the equivalent of thousands of API dollars per month under a flat fee. The $200 credit is a hard cap that reflects API costs, not a continuation of the previous unlimited model. High-volume, continuous agent workloads are no longer viable under a consumer subscription plan. ## Competitive context OpenAI's ChatGPT Plus and Team plans include programmatic Codex usage without a separate metered credit system. OpenAI has also explicitly welcomed third-party tools built on its platform and acquired the team behind OpenClaw. For the period following June 15, OpenAI is offering two free months of service to enterprise customers switching from Claude. Anthropic's direction is toward first-party integration: Claude Code, Claude Routines (a managed agent feature), and official connectors for services like Notion and Google Drive. The programmatic credits system makes third-party tool development financially less attractive and directs programmatic use toward Anthropic's own API or Enterprise tiers. ![ChatGPT subscription tier comparison highlighting that expanded Codex usage is included in the Plus plan](https://imagedelivery.net/xZXo0QFi-1_4Zimer-T0XQ/d25cd689-ab7b-4088-49ba-570d99eba500/lg1x =1920x1080) ## Summary The programmatic credits system resolves the previous ambiguity about what programmatic use is permitted and at what cost. It replaces informal enforcement and silent blocks with a predictable, metered model. **The tradeoff is that the credit values are equivalent to standard API spending, not a subsidized extension of the flat-rate subscription**. For light to moderate programmatic use, the credits provide a clear budget. For the heavy agentic workloads that characterized the previous unlimited model, the new system represents a significant cost increase that pushes that use toward direct API or Enterprise pricing.