From June 15 2026 Anthropic changes how it bills programmatic use of Claude. A separate monthly credit pool of $20 to $200 replaces what was previously included in Pro and Max subscriptions. When the credits run out, requests stop unless you have actively enabled overflow billing. Many Norwegian SMBs will notice this on their next invoice. Here is what happens, who is affected, and what you can do.
What is actually new?
Four surfaces move from subscription to a separate credit pool:
- Claude Agent SDK (Python and TypeScript library for embedded agents)
claude -p(non-interactive headless mode, typically used in cron jobs and scripts)- Claude Code GitHub Actions (CI pipelines)
- Third-party apps that authenticate via your subscription
What does not move: chat on web, desktop, and mobile, plus interactive use of Claude Code in terminal or IDE. If you just chat with Claude, you will not notice anything.
This change connects to a pattern we have already documented: most Norwegian SMBs lack visibility into their own AI usage and costs. When a new pricing model lands without that visibility, the consequences become unpredictable. If you want to understand the full cost picture, read AI strategy for Norwegian SMBs alongside this article.
The credit pools are:
| Plan | Monthly credits |
|---|---|
| Pro | $20 |
| Max 5x | $100 |
| Max 20x | $200 |
| Team Standard | $20 per seat |
| Team Premium | $100 per seat |
| Enterprise (usage-based) | $20 |
| Enterprise Premium (seat) | $200 |
| Enterprise Standard (seat) | $0, no credits |
Notice the last row. Companies on Standard seat agreements get nothing. They must move all automated workloads to a paid API key.
Why is Anthropic doing this?
The background is that subscriptions have subsidized automated use. First-party tools like Claude Code reuse context via prompt cache, while third-party agents often reprocess context. According to Boris Cherny, who leads Claude Code at Anthropic, he said in April that subscriptions are not built for the usage patterns of third-party tools.
Real numbers from community calculations: OpenClaw users on Pro have received about $236 in API value per $20 subscription. Heavy Max 20x users can hit 175x. That is a subsidy that cannot last.
This is the third time this year Anthropic has tried to solve the problem. In January they blocked OAuth tokens for third parties, then reversed within days. In April they banned third-party agents on subscriptions, but reversed within 24 hours after around 135 000 OpenClaw instances were affected. The credit pool model is the compromise that now applies.
Who is affected in Norway?
To understand who is affected, it helps to know what an AI agent actually is, and what it costs to run one. When we talk about "programmatic use" we mean every workflow where Claude runs without a human steering the conversation.
Many Norwegian SMBs sit on Max subscriptions and have built workflows around "unlimited" Claude use. Here are the most common patterns that get hit:
1. CI pipelines with Claude Code GitHub Actions. Many Norwegian dev teams use Claude for code review and test generation in GitHub Actions. Every push draws from the new credits. An active team of 5 developers can burn $200 in under a week.
2. Internal agents with claude -p in cron jobs. Some have set up agents that summarize emails, categorize support cases, or update the CRM. These typically run every hour. Before June this was free under the subscription. Now it counts against the credits.
3. Customer-facing chatbots and assistants. Some SMBs use Claude through third-party tools like Conductor, OpenClaw, or built on Agent SDK. These apps authenticate via your subscription and draw from your credits.
4. Companies on Standard Enterprise seats. This is where the impact is hardest. The credit is $0. Every single automated conversation is billed at API price, with no subsidy.
3 steps to take this week
1. Turn on overflow billing before June 15. Go into your Claude account and activate "usage credits". When the credits run out, requests continue, but you pay API price for the excess. Without this setting, requests stop abruptly. For most Norwegian SMBs this is safer than the alternative, because a failed agent in a customer-facing channel is worse than an extra invoice. But set a cap in the Claude account to avoid surprises.
2. Map what you actually use. Before June 15: log into your Claude account and look at your May usage. How much went to interactive chat, and how much went to Agent SDK, GitHub Actions, and third-party tools? If you already used more in May than the credits allow in June, you know the answer: you must move production workloads to a direct API key.
3. Move shared workloads to an API key. If your team shares a Max 20x credit across multiple users, there is no pool function. Each user's credit is isolated. For shared CI pipelines, scheduled batch jobs, and joint automation, Anthropic itself recommends using Claude Platform with an API key.
In practice that means you set up a separate API key with a spend cap, and point the workload at it. The billing then stays flat and predictable each month, regardless of how much your developers chat with Claude.
Need help mapping what you use today and what should move? Our AI implementation guide describes the methodology we use to move customers from experiment to production.
What does it cost in practice?
API price for Claude Sonnet 4 is $3 per million input tokens and $15 per million output tokens. A typical Norwegian SMB team of 5 developers running Claude-driven code review in GitHub Actions can easily use 50 to 100 million tokens per month. That is $300 to $600 per month after the credits run out.
Compared to a Max 20x subscription at $200, this is a real cost increase for those who have become used to unlimited usage. But it is also a reminder that AI tokens are an operational cost. They should be budgeted and measured, not consumed freely.
How can AIKI help?
We help Norwegian SMBs understand what they actually use, and what pays off to move. Three concrete things:
- AI cost analysis: we go through subscriptions, API use, and third-party tools. The service is called AI audit.
- Migration to API key: we set up a separate Claude account for production workloads, with spend cap and logging.
- AI governance for SMB: we build a cost dashboard so you see consumption per workflow.
Book an AI kickstart to figure out whether you should stay on subscription, move to API key, or do a mix. We have also written about how we build cost dashboards for AI usage.
In summary
June 15 2026 is a real change for Norwegian SMBs using Claude. The subscription still covers chat and interactive use, but programmatic use gets its own credit pool of $20 to $200. When the credits run out, requests stop unless you have enabled overflow billing. Production workloads should move to API key for predictability. The biggest risk is not the invoice, it is that agents stop without warning. Set up monitoring and spend cap before June 15.
Need help mapping AI costs and moving production workloads? Book an AI kickstart or read about AI audit.
