Anthropic's Max and Team API credits: how they change agent experiments

Many teams that use Claude have two separate bills: a subscription for people using Claude and Claude Code interactively, and a metered Claude Platform account for anything built on the API. On 7 October 2026 Anthropic started to connect the two. Claude Max and Team subscribers now receive a monthly credit for the Claude Platform, aimed at building tools, apps and agents that call the API. The amounts are modest, the rules are specific, and the most important one is easy to miss: the credits do not pay for interactive Claude Code.

What Anthropic announced

The credits were announced alongside Claude Haiku 5.5. Anthropic says they are designed to let subscribers "experiment with building tools, apps, and agents that call our API", and that they can be used on any of its models. (What Haiku 5.5 itself changes for coding sub-agents is covered in Claude Haiku 5.5 for coding sub-agents.) The Help Center article and the Claude Platform documentation give the detail.

Plan	Monthly credit
Max 5x	$100
Max 20x	$200
Team, per Standard seat	$20, pooled
Team, per Premium seat	$100, pooled

On Team plans, including discounted Nonprofit and Scientists plans, the seat credits are pooled into one balance capped at $500 a month. Both pages give the same example: three Standard seats and two Premium seats receive $260 a month, and one more Premium seat raises the next cycle's credit to $360. Free, Pro and Enterprise plans are not eligible.

The Help Center says the credits are being rolled out over a few days, so some eligible subscribers may not see the offer yet.

How claiming works

A subscriber must have been on an eligible plan for seven days. On Max, the subscriber claims; on Team, a Primary Owner or Owner does. They also need the Owner, Admin or Billing role in a Claude Console organisation, which receives the credits. No payment method is required on the Claude Platform.

Two constraints matter for planning:

• One organisation, fixed. Each plan links to one Console organisation, and each organisation can receive credits from only one plan. The linked organisation cannot be changed without contacting support.
• Shared balance. Everyone with an API key in the linked organisation draws from the same balance. To limit a project or a person, give them a workspace with its own spend limit.

What the credits cover, and what they do not

The Platform documentation lists the Claude API, Claude Managed Agents, the Claude Agent SDK and the Playground as covered, and suggests the Message Batches API to make the credits go further. Not covered:

• Claude Code, which the Help Center specifies as interactive Claude Code in the terminal, IDE, desktop or web.
• Extra usage in the Claude apps.
• Claude on other clouds: the documentation lists the Claude Platform on AWS, Amazon Bedrock, Google Cloud and Microsoft Foundry as excluded.

There is a nuance that the two pages express differently. The Platform documentation's table simply says Claude Code: No. The Help Center adds that running claude -p or the Agent SDK yourself with an API key from the linked organisation is covered, because that usage is billed as Agent SDK usage. When signed in with the Claude plan instead, claude -p draws from plan limits, and runs started by the Claude Code GitHub Action, an IDE extension or the desktop app count as Claude Code usage and are not covered, even with -p. Teams planning to use the credits for scripted agent runs should follow the Help Center's more detailed description and test it.

The Help Center also clarifies that this is not the Agent SDK credit announced in June, which it says is not available.

How the balance behaves

• No rollover. Unused credits expire at the end of each billing cycle.
• Used first. Included credits are spent before purchased credits. Auto-reload is based only on the purchased balance, so it can trigger even while included credits remain.
• Hard stop by default. If the organisation has no other credits, API requests stop until the next monthly credit arrives. Usage is never charged to the Claude plan. Organisations invoiced through Anthropic sales are billed as usual beyond the credits.
• Separate from plan limits. The credits do not change usage limits in Claude or Claude Code.
• Spend caps and credits run on different clocks. Usage paid with credits counts towards the organisation's monthly spend cap, which resets on the first of each calendar month, while credits refresh on the billing cycle.
• Tier changes. Linked organisations move up to at least the Start usage tier, or at least the Build tier if their monthly credits exceed $200. Otherwise the credits do not count towards higher tiers.

What it means for engineering and security leaders

For a team that already pays for Max or Team seats, the credits are a small, free budget for building on the API: an internal review bot, a batch job that summarises incidents, a prototype agent on Managed Agents. They are not a discount on the interactive coding traffic that makes up most subscription usage, and they should not be treated as one in forecasts.

They also create a governance question. Credits land in one Console organisation that cannot easily be changed, and everyone with a key in that organisation shares them. If the linked organisation is also where production API traffic runs, the credits will be consumed by production first, since included credits are spent before purchased ones. For many teams, a separate organisation for experiments, with workspace spend limits, keeps the accounting clear. The article on subscription and metered AI pricing covers the wider trade-off.

The hard stop is useful for experiments, because a runaway agent cannot run up a bill. It is a problem for anything people rely on, which will simply stop when the credit is spent if no other credits are configured.

Practical takeaways

• Decide which Console organisation to link before claiming, since it cannot be changed without support.
• Prefer a dedicated organisation for experiments, so production traffic does not consume the credits.
• Set workspace spend limits per project or person within the linked organisation.
• Do not count the credits against interactive Claude Code costs in forecasts.
• If you plan to use claude -p with the credits, run it with an API key from the linked organisation and confirm the billing.
• Remember that spend caps reset on the calendar month while credits follow the billing cycle.

Where SpanDock fits

The SpanDock model proxy is relevant if you route agents through it: only API-key traffic is routed, and Claude Pro and Max sign-ins are not used while the proxy is on. Calls that go through it record the model, tokens and cost, and per-person budgets are available on the Standard plan and up. The article on per-person AI budgets and the model advisor explains how those fit together.