Anthropic's new monthly API credit for Max and Team: what it covers, what it doesn't, and which pool your claude -p runs draw from

The 30-second version

On 7 October, in the same post that launched Haiku 5.5, Anthropic said it would “roll out a new monthly API credit to all Max and Team subscribers for use on the Claude Platform”: $100 a month on Max 5x, $200 on Max 20x, and up to $500 on Team, pooled across seats. The support page and the platform documentation that go with it answer most of the questions people asked in the first hour. The short answers:

• It is a Console balance, not a plan top-up. You claim it by linking one Claude Console organization to your plan (you can create one in the flow; no card needed). The credit lands in that organization each billing cycle — monthly on annual plans — is spent before any credit you bought, is shared by every API key in the organization, and expires at the end of the cycle. Your plan’s usage limits do not change.
• It covers building, not Claude Code. Messages API, Batches API, the Playground, Claude Managed Agents and the Claude Agent SDK, on any current model: yes. Interactive Claude Code in the terminal, IDE, desktop or web: no. Extra usage in the Claude apps: no. Bedrock, Vertex and Foundry: no.
• claude -p is covered only through an API key. With a key from the linked organization, a claude -p run is billed as Agent SDK usage and draws on the credit. Signed in with your plan, the same command draws on plan limits as before. Runs started by the GitHub Action, an IDE extension or the desktop app count as Claude Code even with -p.
• The credential decides, and in our checks the key won. On Claude Code 2.1.293 and 2.1.294, with no higher-priority credential set, ANTHROPIC_API_KEY switched claude auth status from claude.ai to api_key, and a claude -p run with a deliberately invalid key failed with a 401 rather than falling back to the plan login. Interactive sessions ask once.
• It is not the June plan. In May, Anthropic said programmatic usage would get its own monthly budget from 15 June, with plan limits reserved for interactive use; it paused that on the day. The October credit leaves plan usage alone, lives in a Console organization, needs an API key, and leaves Pro out.
• $200 goes a long way for scripted work. At the list-price costs Claude Code reported for our runs, it buys 266 Opus 5.5 runs of Claude Code’s prompt audit or 5,154 on Haiku 5.5; the measurements behind our last four articles cost $9.78 together. It would not cover the subscription-backed traffic one of our machines recorded, which prices out at roughly $900 a month and is mostly interactive, which the credit excludes.

Verdict: claim it if you have, or would like, a Console organization for scripted work — batch jobs, evaluation harnesses, agents built on the SDK — and keep interactive sessions on the plan. Set the API key per command or per project rather than in your shell profile, or every claude -p run that inherits that environment quietly leaves the plan.

What changed on 7 October, and what it replaced

This is the second time this year Anthropic has put a monthly credit next to the Max plan, and the two designs are different enough that the first one still confuses the second. We read the 13 May thread from Anthropic’s developer account directly:

The 15 June pause note is still on the Agent SDK support page: “For now, nothing has changed: Claude Agent SDK, claude -p, and third-party app usage still draw from your subscription limits.” The 7 October note above it adds that Max and Team plans now include monthly API credits “which cover the Claude Agent SDK, claude -p, the Claude API, and Claude Managed Agents,” and that “you can still use the Claude Agent SDK, claude -p, and third-party apps with your subscription limits.” Read together: nothing was taken away, and a second pool was added for those with an API key.

The early Hacker News reaction read the second sentence as the first — one post was titled “Claude Max and Teams plans now forced to API” — because the June plan had primed people to expect programmatic usage to leave the plan. It has not. A commenter quoting the archived page noted that the same support article carried the June wording until 6 October, which explains the confusion.

Whether the new design is better depends on who you are. A Max subscriber now has both pools, where the June plan would have split one into two. A Pro subscriber, who would have had $20, gets nothing. And anyone whose scripts run signed in with the plan gets nothing new until they move them to an API key.

What the credit covers and what it doesn’t

Straight from the support page and the platform documentation, with the two details that matter most for Claude Code users in the last rows:

The platform page spells out the failure mode for the obvious workaround. If a Console organization holds only these credits and you point a Claude Code session at it, Claude Code shows Credit balance too low · Add funds, with a link to the Console billing page. The credit is not a second plan.

One naming trap. Claude Code’s own documentation uses Usage credits, formerly “extra usage,” for the overflow a Pro or Max subscriber can turn on under Settings, Usage to keep working past plan limits at API rates against a spend limit. API credits are this new Console balance. Different places, different things, and the new one cannot be spent as the old one.

Which pool a claude -p run draws from: checked

Anthropic’s rule is that the credential decides, so we checked how Claude Code chooses when more than one is available. The authentication documentation lists the order: cloud-provider credentials when a Bedrock, Vertex or Foundry variable is set; then ANTHROPIC_AUTH_TOKEN; then ANTHROPIC_API_KEY; then an apiKeyHelper script; then CLAUDE_CODE_OAUTH_TOKEN; then Anthropic profile credentials; and only then the subscription login from /login. In interactive mode, Claude Code asks once whether to use an ANTHROPIC_API_KEY it finds and remembers the answer (the “Use custom API key” toggle in /config appears only while the variable is set). In non-interactive mode, the documentation says, “the key is always used when present.”

We ran the checks from a clean environment on a Max 20x subscription, with none of the higher-priority credentials set:

• claude auth status shows the method. Signed in with the plan it reports "authMethod": "claude.ai" and "subscriptionType": "max". With ANTHROPIC_API_KEY set it reports "authMethod": "api_key" and "apiKeySource": "ANTHROPIC_API_KEY", and the subscription fields disappear. Checked on Claude Code 2.1.293 on 7 October and 2.1.294 on 12 October. It shows which credential will be sent, not which balance the key’s organization will bill — the Console’s billing page is where you confirm that the organization holds the credit.
• In -p mode the key won. With a deliberately invalid key in the environment and a valid plan login in the keychain, claude -p returned api_error_status: 401 and no cost on both dates — “Failed to authenticate. API Error: 401 API key is invalid” on 2.1.293, “Invalid API key · Fix external API key” on 2.1.294. It did not fall back to the plan.
• An environment OAuth token also beat the stored login. On 12 October only, a deliberately invalid CLAUDE_CODE_OAUTH_TOKEN produced “401 OAuth access token is invalid,” matching the documented order.

These checks show which credential Claude Code sends, not what a valid key is billed to; the support page is the source for that side — an API-key claude -p run is Agent SDK usage, a signed-in one is plan usage.

The practical consequence is the trap in the lead. A key exported in .zshrc for some other tool makes every claude -p run in a shell that inherits it an API-key run — billed to that key’s organization, on the credit while it lasts, then to purchased credits or, for a prepaid organization with nothing else, not at all. According to the documentation, interactive sessions ask once and remember the choice; change it through “Use custom API key” in /config. Claude Code 2.1.296’s changelog says its refused-session errors now “name an API key or auth token that overrides your claude.ai login,” which is the same hazard seen from the other side. The clean pattern is a per-command or per-project key: ANTHROPIC_API_KEY=… claude -p for the jobs you want on the credit, and nothing in the shell profile.

Third-party apps are a separate case. The Agent SDK documentation states that, unless previously approved, third-party developers may not offer claude.ai login or plan rate limits in their products, and the support page says such apps can still run on your subscription limits. An app that takes an API key from your linked organization draws on the credit for the requests the credit covers; one that signs in with your plan draws on the plan. We did not test any.

What $100 or $200 buys at list price

The best yardstick we have is the work we have already measured. These are claude -p jobs on small repositories, the shape the credit is meant for, at the list-price costs Claude Code reported. Counts are complete repetitions:

For contrast, the traffic a Max subscriber would most like to offload is interactive Claude Code, and that is exactly what the credit excludes. The 12,901 subscription-backed calls we scanned from one machine between 10 July and 7 October price out at $2,678 on Opus 5.5’s sheet, roughly $900 a month. That is a modeled figure, since those sessions ran on a subscription, and we did not classify every call by execution mode; nearly all of them came from interactive sessions and their subagents, which the credit does not cover. The $200 credit is about a fifth of that figure and could not have been spent on most of it. The measurements listed in the table total $9.78 at list prices, so $200 represents about twenty repetitions of that particular workload.

Six details in the fine print

1. One organization, chosen once. Each plan links to one Console organization and each organization can receive credits from one plan. You cannot change the link yourself; changing it later means contacting support. Pick the organization you will actually build in.
2. Auto-reload ignores the free balance. Included credits are spent before purchased credits, but auto-reload watches only the purchased balance and “reloads when that balance reaches your threshold, even if included credits remain.” If auto-reload is on, you can buy credits while free ones are still unspent.
3. It counts toward your spend cap. Usage paid with credits counts toward the organization’s monthly spend cap, which resets on the first of the month while the credit refreshes on your billing cycle. The reset dates may differ.
4. When it runs out, a prepaid organization stops. If the organization has no purchased credits and no auto-reload, API requests return “Your credit balance is too low to access the Anthropic API” until the next monthly credit arrives. Usage is never charged to the Claude plan. Organizations invoiced through Anthropic sales are the exception: their usage beyond the credit is billed as usual.
5. The rate-limit tier moves a little. A linked organization is placed at least in the Start tier, and at least in Build if its monthly credit exceeds $200 (so Team pools above $200). The credit does not otherwise count toward tier progression.
6. Timing follows your plan. New subscribers see the offer after seven days on an eligible plan; credits arrive shortly after each plan payment, monthly on annual plans; an upgrade from Max 5x to 20x prorates immediately (later, by the next renewal, for Google Play subscriptions); cancelling stops new credits but leaves the current ones usable until they expire.

On our own Max 20x account the offer appeared on 7 October as a “Free API credits” card under Settings, Billing, with an API credits section showing $200. The documented next step is linking an organization and accepting the credit terms. We left it unclaimed for this article: we report what the pages say and what Claude Code does, not a claim flow we could not undo.

What to do

1. Claim it if scripted work is part of your month. Evaluation harnesses, batch rewrites, scheduled claude -p jobs, anything built on the Agent SDK. Create or choose the Console organization deliberately, since the link is permanent from your side.
2. Keep interactive work on the plan. The credit cannot pay for it, and the documentation says a credits-only organization fails a Claude Code session with “Credit balance too low.” Your five-hour and weekly limits are unchanged either way.
3. Scope the key. Use ANTHROPIC_API_KEY=… claude -p … for the runs you want on the credit, or a project-level env block in a settings file, and keep the key out of your shell profile. claude auth status shows which credential a shell will send; the Console shows whether that key’s organization holds the credit.
4. Give credit-funded work its own workspace. Every key in the organization draws from one balance; a workspace with a spend limit keeps one runaway script from emptying the month’s credit for everything else.
5. Check auto-reload before you rely on the free balance. If it is on, decide whether you want it topping up purchased credits while the included ones are still there.
6. Don’t read it as the June change. Nothing moved off the plan. If you rebuilt a workflow in May to avoid the planned split, the October credit is a reason to script more, not less.

What we didn’t test

We did not claim the credit. So we did not see the Console balance, the “Promotional credits” line, the expiry date, or what happens at exhaustion; those descriptions come from Anthropic’s pages. Linking an organization is a one-way change to an account we use for other measurements.

Team pooling, Google Play timing, the invoiced-organization exception and the rate-limit tier bump are reported as documented, not observed.

Third-party apps. We did not test Zed, Paseo, T3 Code or any other SDK-based tool against a linked organization. What we say about them comes from the SDK documentation’s authentication rule and the support page’s statement that they can still use subscription limits.

Our precedence checks used invalid credentials on purpose, in an environment with no higher-priority credential set. They show which credential Claude Code sends, not what a valid key from a linked organization is billed to, and a cloud-provider variable, ANTHROPIC_AUTH_TOKEN or an apiKeyHelper script would rank above the key we tested.

The “what it buys” table is list price. It reuses costs Claude Code reported for runs that actually drew on a Max subscription; on a Console key they would be billed at the same list rates, prompt caching included.

Companion reading

• How Claude Code’s usage limits actually work — the five-hour window, weekly caps and Usage credits the new credit sits beside, not inside.
• Haiku 5.5’s 90% price cut stops at 100K tokens — the per-run costs reused above, and the traffic the credit does not cover.
• Claude Code’s /doctor prompt-audit, tested on nine planted problems — the Opus and Sonnet runs priced in the table.
• Codex vs Claude Code usage limits — how the two plans meter subscription usage.
• Claude Code pricing in 2026 — which plan tier pays off, now with $100 or $200 of API credit on the Max side of the ledger.

Sources

1. Anthropic — Introducing Claude Haiku 5.5, 7 October 2026. “Second, this week, we’ll roll out a new monthly API credit to all Max and Team subscribers for use on the Claude Platform.” Amounts per plan; “designed to allow our users to experiment with building tools, apps, and agents that call our API.”
2. Anthropic — Monthly API credits for Max and Team plans, support article, retrieved 12 October 2026 and marked “Updated this week.” Eligibility, the claim flow, what the credits cover, and the FAQ on claude -p, the GitHub Action, IDE and desktop runs, and the June credit.
3. Anthropic — API credits for Max and Team plans, Claude Platform documentation. The at-a-glance table, auto-reload behavior, spend caps, rate-limit tiers, the invoiced-organization exception, the exhaustion messages, and Credit balance too low · Add funds for uncovered requests.
4. Anthropic — Use the Claude Agent SDK with your Claude plan, support article with the 15 June pause note and the 7 October update.
5. Claude Developers on X — thread of 13 May 2026, read on 12 October 2026: “Starting June 15, paid Claude plans can claim a dedicated monthly credit for programmatic usage,” the amounts per plan, “Your subscription limits don’t change, they’re now reserved for interactive use,” overflow via usage credits, and third-party tools drawing from the credit.
6. Claude Developers on X — post of 7 October 2026 announcing monthly API credits for Max and Team subscriptions.
7. Anthropic — Claude Code: Authentication. The credential precedence list, the one-time prompt in interactive mode, and “In non-interactive mode (-p), the key is always used when present.”
8. Anthropic — Claude Code changelog, 2.1.296: refused-session errors “name an API key or auth token that overrides your claude.ai login.”
9. Anthropic — Agent SDK overview. Third-party developers may not offer claude.ai login or rate limits unless approved; use API-key authentication.
10. Anthropic — Claude Code: Manage costs effectively. Usage credits, the /usage-credits command, and Settings, Usage on claude.ai.
11. Hacker News — Claude Agents SDK will no longer use subscription; API credits included in plans, 7 October 2026, including a commenter’s quotation of the support page’s 6 October wording.

Our own measurements: claude auth status and claude -p runs from a clean environment on macOS, on a Max 20x subscription, with Claude Code 2.1.293 on 7 October and 2.1.294 on 12 October 2026, using a deliberately invalid ANTHROPIC_API_KEY on both dates and a deliberately invalid CLAUDE_CODE_OAUTH_TOKEN on 12 October; list-price costs from the runs reported in our 26 September, 4 October and 7 October articles.

FAQ

Does the Max plan’s monthly API credit cover Claude Code? No. Anthropic’s terms exclude interactive Claude Code in the terminal, IDE, desktop and web, and the credits do not change your plan’s usage limits. The credit covers the Claude API (Messages and Batches), the Playground, Claude Managed Agents and the Claude Agent SDK, used with an API key from the Console organization you link. Anthropic’s documentation says that if a Console organization holding only these credits is used for a Claude Code session, Claude Code shows “Credit balance too low.”

Does the API credit cover claude -p? Only when claude -p runs with an API key from your linked Console organization, because that usage is billed as Agent SDK usage. When you are signed in with your Claude plan, claude -p and Agent SDK usage still draw from the plan’s usage limits. Runs started by the Claude Code GitHub Action, an IDE extension or the desktop app count as Claude Code and are not covered even with -p.

How do I know which credential a Claude Code run is using? Run claude auth status. Signed in with a plan it reports authMethod claude.ai and your subscription type; with ANTHROPIC_API_KEY set it reports authMethod api_key and apiKeySource ANTHROPIC_API_KEY. That tells you the authentication method, not which balance the key’s organization will bill; check the organization and its credit in the Console separately. In non-interactive mode the documentation says the key is always used when present, and in our checks on Claude Code 2.1.293 and 2.1.294, with no higher-priority credential set, a run with a deliberately invalid key failed with a 401 instead of falling back to the plan login. Interactive sessions ask once and remember the answer.

Is this the Agent SDK credit Anthropic announced for June 15? No. On 13 May, Anthropic’s developer account announced that from 15 June programmatic usage — the Agent SDK, claude -p, the GitHub Action and third-party apps — would get its own monthly budget, with plan limits reserved for interactive use, at $20 on Pro, $100 on Max 5x and $200 on Max 20x. Anthropic paused that change on 15 June, and the support page now says that credit isn’t available. The October credit is a different design: plan usage stays as it was, the credit lives in a Console organization and is spent through API keys, and Pro plans are not included.

What is the difference between API credits and Usage credits? Usage credits, formerly called extra usage, let a Pro or Max subscriber keep working past the plan’s limits, billed at API rates against a spend limit set under Settings, Usage. API credits are the new monthly balance granted to a Claude Console organization for building on the API. They sit in different places, pay for different things, and the API credit cannot be used as extra usage for Claude Code.

How much is $200 of API credit in practice? At the list-price costs Claude Code reported for our own runs: 266 runs of Claude Code’s prompt audit on Opus 5.5 at $0.75 each, 487 on Sonnet 5.5 at $0.41, or 5,154 on Haiku 5.5 at $0.0388. The 48-run mod permission matrix from our 4 October test cost $0.92, so the credit covers it 217 times; the measurements behind our last four articles came to $9.78 together, about 20 repetitions. It would not cover the subscription-backed Claude Code traffic one of our machines recorded, which prices out at roughly $900 a month on Opus 5.5 and is mostly interactive, which the credit excludes.