Anthropic Introduces Dedicated Credit Pools for Agent SDK Usage
Introduction
Anthropic has announced a significant change to how Claude subscription plans handle programmatic usage. Starting June 15, all usage originating from the Agent SDK—including third-party apps built on it—will be billed from a new, separate monthly credit pool. This move creates a clear distinction between interactive human chats and automated, code‑driven interactions.

What’s Changing
In a post on X (formerly Twitter) on Wednesday, Anthropic revealed that paid Claude subscribers will soon receive a monthly Agent SDK credit. This credit covers programmatic usage: the Claude Agent SDK itself, the claude -p command, Claude Code GitHub Actions, and any third-party application authenticated via the Agent SDK.
Programmatic vs Interactive Usage
The new policy draws a firm line between two types of usage:
- Interactive usage—chatting with Claude on the web, desktop, or mobile; using Claude Code in the terminal; or working with Claude Cowork—will continue to count against your standard subscription limits.
- Programmatic usage—invoking
claude -pin Claude Code, running SDK scripts in Python or TypeScript, or letting third-party tools authenticate through the Agent SDK—will consume your new monthly Agent SDK credit.
Previously, both types of usage drew from the same pool, so subscribers didn’t have to worry about the distinction. Now, the separation requires a bit more attention, especially for heavy automation users.
How the New Credit System Works
As Anthropic outlined on X, Claude users must claim a separate monthly credit to enable Agent SDK usage. The claim needs to happen only once; after that, all programmatic usage will automatically draw from that credit pool.
If You Exhaust Your Credit
What happens when your Agent SDK credit runs out? You have two options:
- Turn on usage credits—you can continue working without interruption, but your ongoing usage will be billed at pay‑as‑you‑go API rates.
- Turn off usage credits—your programmatic usage will be capped for the month. You’ll have to wait until the next billing cycle for your credit to refresh.
Unused Credits Do Not Roll Over
Anthropic has confirmed that any leftover Agent SDK credit at the end of a billing cycle will expire. There is no rollover, so it’s advisable to plan your usage accordingly.
Credit Amounts by Plan
The size of your Agent SDK credit pool depends on which Claude plan you subscribe to. Here’s a breakdown:

Individual Plans
- Pro: $20 credit
- Max 5x: $100 credit
- Max 20x: $200 credit
Team and Enterprise Plans
- Standard Teams: $20 per seat
- Premium Teams: $100 per seat
- Enterprise Usage‑Based: $20
- Enterprise Premium Seats: $200 per seat
It’s important to note that Agent SDK credits are assigned to individual accounts, not shared across a team. You cannot transfer or pool credits. Additionally, accounts that use the Claude Developer Platform with an API key are not eligible for these credits—they will continue under pay‑as‑you‑go billing.
Implications for Third‑Party Tools
On X, Anthropic’s developer communications account clarified how this affects third‑party applications built on the Agent SDK: “Third‑party tools built on the Agent SDK like Conductor and OpenClaw work with your Claude plan, but will draw from your credit the same way your own scripts do.” In other words, if you rely on external apps that integrate with Claude via the Agent SDK, their usage will now be deducted from your dedicated credit pool, just like your own automated scripts.
Conclusion
This change by Anthropic aims to provide clearer billing boundaries between interactive and programmatic usage, giving power users more control over how they allocate their subscription benefits. While the introduction of a separate credit pool may require some adjustment, it also offers transparency and prevents automation from eating into your manual chat quota. As the June 15 deadline approaches, Claude users should review their typical usage patterns to decide how best to manage their new Agent SDK credits.
Related Articles
- From Underdog to Leader: How Anthropic's Claude Overtook OpenAI in Enterprise AI Adoption – and What to Watch Out For
- How to Train Multiple LLM Sizes Simultaneously with NVIDIA Star Elastic
- Smart LLM Routing on a Budget: A NadirClaw Q&A Guide
- How to Deploy AWS's Latest Agentic AI Tools for Your Business
- OpenAI Launches Next-Generation Voice Models for Real-Time Audio Applications
- Unlock Claude’s Built-in Project Manager: A Step-by-Step Guide
- Mastering Claude Opus 4.7 on Amazon Bedrock: A Complete Deployment Guide
- The Hidden Cost of AI Friendliness: 7 Critical Facts from Oxford Research