ChatGPT subscription API credits: $100 to $500 a month, and why they will not pay for Codex or Claude Code

ChatGPT subscription API credits are now worth $100, $200 or $500 a month: OpenAI's developer docs grant that much to its three paid Pro tiers, and $20 per Standard seat or $100 per Premium seat on Business, pooled and capped at $500 per workspace. Anthropic grants $100 on Max, $200 on Ultra and the same pooled seat amounts on Team. Every grant expires at the end of your billing cycle.

That changes the build-versus-subscribe math for small teams, but only if you read three rules first: the credits expire with no rollover, they land in exactly one API organization you cannot re-point yourself, and neither vendor lets them pay for its coding agent. Here is what each vendor's own documentation says, as of October 2026.

What ChatGPT subscription API credits are worth, plan by plan

On the OpenAI side the credit tracks your Pro tier. The developer documentation lists the $100, $200 and $500 Pro tiers receiving $100, $200 and $500 a month respectively, issued automatically on renewal, with the first grant prorated for the rest of the current cycle. The tier labels are OpenAI's own.

ChatGPT Business pools instead: $20 for each paid Standard seat and $100 for each paid Premium seat, capped at $500 a month per workspace. Workspaces that pay annually up front receive the whole year's grant at once, and it expires at the end of that annual cycle. Credits added by mid-cycle seat changes are capped at $6,000 per annual cycle. Seat reductions take effect next cycle, and reassigning a seat creates no new credit.

Plus, Go, Free and ChatGPT Enterprise are not eligible. Neither are free trials or complimentary subscriptions.

Claude Max, Ultra and Team credits equal the plan price

Anthropic's API credits for Max, Ultra and Team plans page sets the credit at the plan price: $100 on Max, $200 on Ultra, $20 per Standard Team seat and $100 per Premium Team seat, pooled into one balance capped at $500. Discounted Nonprofit and Scientist Team plans get the same per-seat amounts. The docs give the worked example: three Standard seats and two Premium seats receive $260 a month, and adding one more Premium seat raises next month's credit to $360.

The match with the subscription price is exact only on annual billing. Anthropic's pricing page lists a Team Standard seat at $20 a month billed annually but $25 billed monthly, while the credit stays $20 a seat either way. Max is $100 a month and Ultra $200, so those two line up on any billing term. Free, Pro and Enterprise plans are not eligible.

The credits do not cover Claude Code or Codex through API keys

This is the detail that catches people who assume a subscription credit pays for the agent they already use. Anthropic's docs are explicit: the credits cover the Claude API, Claude Managed Agents, the Claude Agent SDK and the Console playground, and they do not cover Claude Code, extra usage inside the Claude apps, or Claude models on AWS, Bedrock, Google Cloud and Microsoft Foundry. Send a Claude Code session from an organization holding only these credits and the docs say you get "Credit balance too low".

OpenAI draws a similar line. Its docs say Codex usage through API keys is not supported, and that the credits do not increase your ChatGPT or signed-in Codex limits. What they do cover is broad: the Responses API, the Agents API including OpenAI-hosted sandbox charges, the Decisions API, GPT-Live and Realtime, the Images API, and built-in tools such as web search, file search and Code Interpreter with their usage and storage charges.

So the credit funds the applications and agents you build on the API. It does not subsidize the subscription-side agents you already pay for, on either platform. We list 246 AI coding tools as of October 2026, so check how each one bills: a tool that calls the API with your own key draws on the credit, while a tool with its own seat price does not.

Expiry, one organization, and the seven-day wait

Both vendors require seven days on an active paid plan before you can claim. Both apply the credits to a single API organization, and both warn that you cannot re-point the link yourself: Anthropic says each plan links to one Console organization and each organization accepts credits from one plan, changeable only by contacting support, and OpenAI says the same about its API organizations. OpenAI additionally requires that you own both the subscription and the organization and that both accounts use the same email address. On Anthropic's side you need to be the subscriber, or a Team Primary Owner or Owner, plus Owner, Admin or Billing in the Console.

Unused credits expire at the end of each billing cycle with no rollover, on both platforms. Included credits are spent before any credits you bought, so a purchased balance is not consumed while a grant is live. Anthropic adds that auto-reload watches only the purchased balance.

When the balance empties, OpenAI's API returns HTTP 429 with the error code credit_balance_exhausted. Anthropic's stops serving requests until the next monthly grant arrives and says usage is never charged to your Claude plan. Anthropic also notes two side effects worth knowing: credit-funded usage still counts toward your organization's monthly spend cap, and a linked organization moves up to at least the Start tier, or the Build tier if its monthly credits exceed $200. Under Anthropic's credit terms, credits are non-transferable and have no cash value.

Is subscribing cheaper than paying for the API? Only up to the credit

The credit is spent at ordinary token prices, so work out what your workload costs before treating the grant as a budget. Claude Sonnet 5.5 is $2 per million input tokens and $10 per million output, which turns a $100 Max grant into roughly 50 million input tokens, or 10 million output tokens, or double either figure through the Message Batches API at half price. OpenAI's pricing page puts gpt-6.1-sol at the same $2 and $10, and gpt-6-luna at $0.10 and $0.50, with Batch again at half.

Past the credit you pay list price like everyone else. Skip this as a production plan if your traffic is steady and uptime matters: requests stop dead when a grant runs out unless you buy credits or turn on auto-reload, and an unused grant does not bank for a busy month. It is a real discount for prototypes, internal tools and spiky agent work, not a free tier. PricePerToken is one way to check where a model sits against rivals before you commit, and our LLMs and AI infrastructure category lists 275 platforms as of October 2026 if the answer is to run the workload somewhere else.

What to do before your next billing date

1. Confirm your plan qualifies: Pro or Business on ChatGPT, Max, Ultra or Team on Claude, paid and at least seven days old.
2. Decide which API organization should hold the credits before you link, because support has to undo it.
3. Claim it in ChatGPT Settings then Billing, or on claude.ai under Settings then Billing, or Organization settings then Billing for a team.
4. Move anything the credit will not pay for, Claude Code and Codex via API keys, back onto its own budget line.
5. Give each project its own workspace with a spend limit so one job cannot drain the shared grant.

Simon Willison documented both launches on October 29, 2026, and the vendor pages above are the versions we checked on October 31. Both vendors can change the program, so re-read the terms before you build a budget on them. If you ship something on the credits, list it on Upfynd and let people vote on it.