OpenAI API credits for ChatGPT Pro vs Claude Max API credits: matching numbers, a different fence around the coding agent

On October 28 OpenAI started giving ChatGPT Pro subscribers API credit equal to their plan: $100, $200 or $500 a month, depending on the Pro tier. Business workspaces get $20 per Standard seat and $100 per Premium seat, pooled and capped at $500. Three weeks earlier Anthropic did the same for Claude Max and Team, at a time when OpenAI’s only subscriber credits were ChatGPT usage credits its help page described as “not API credits.” That sentence is now out of date.

Simon Willison, who keeps subscriptions with both, called it “pretty much exactly the same deal.” On the seat plans it is: Team and Business use identical numbers and the same $500 cap. On individual plans Claude Max stops at $200, and only OpenAI has a $500 tier.

Two differences matter more than the amounts. Both labs keep the credit away from their own coding agent, but Anthropic draws the line by type of usage and OpenAI by product, so a headless Claude script can spend the credit and a Codex run on an API key can’t. And each lab documents one billing rule the other leaves unstated.

How do OpenAI’s ChatGPT API credits and Claude’s Max API credits work?

Almost identically. A subscription links to one API organization, an organization accepts one subscription, and changing the link means contacting support. Both require seven days on a paid plan before you can claim. Credit expires at the end of each billing cycle with no rollover (Claude’s annual plans still get and lose it monthly), the free credit is spent before anything you purchased, and none of it raises your chat or coding-agent limits. Neither needs a card on the API side.

The quoted amounts are full-month figures. OpenAI prorates the first grant for the time left in your billing period, so a claim made late in the cycle shows less.

Does the OpenAI API credit cover Codex, and does Claude’s cover Claude Code?

OpenAI’s doesn’t cover Codex on an API key at all. Its help page is flat:

This benefit does not cover Codex usage through API keys. These API
credits also do not increase your ChatGPT or signed-in Codex usage limits.

Anthropic’s excludes interactive Claude Code and runs started from the GitHub Action, IDE extensions and the desktop app. But its help article says the credit does cover claude -p and the Agent SDK when you run them yourself with a key from the linked organization, because that usage is billed as Agent SDK usage.

So a headless coding script you wrote can spend Anthropic’s credit. On OpenAI’s side, the plain reading is that codex exec authenticated with an API key is out, and Codex stays on your signed-in ChatGPT limits or on a paid API balance. If you split work between Claude Code and Codex, the two credits fund different halves of the setup.

What does the OpenAI API credit cover that Claude’s pages leave open?

Hosted tools and sandboxes, in writing. OpenAI’s developer guide lists the Agents API “including OpenAI-hosted sandbox charges,” and built-in tools such as web search, file search and Code Interpreter “and their associated usage and storage charges.” Responses, Chat Completions, the Decisions API, Realtime and Images are covered too.

Anthropic covers the Claude API, Managed Agents, the Agent SDK and the Console playground, on any model. As of October 9, its pages don’t say whether server-side web search or Managed Agents runtime is paid from the credit. It runs through the same organization, so it may well be, but only OpenAI has written it down.

How do refunds, seats, annual plans and linking rights differ for OpenAI and Claude API credits?

Each of these can move real money:

• Refunds. On OpenAI, an approved full refund removes the unused promotional credit tied to that charge. On Anthropic, credits you already have stay usable until they expire, refund included.
• Seats. Adding a paid Business seat on OpenAI can produce extra prorated credit within the cycle, up to the cap. On Claude Team, each month’s credit uses the seats present at the start of that billing month.
• Annual billing. An OpenAI Business workspace paying annually gets the whole year at once, capped at $6,000, expiring at the end of the annual cycle.
• Who can link. OpenAI requires you to own both the subscription and the API organization, with the same email; an admin role isn’t enough. Anthropic accepts Owner, Admin or Billing in the Console.

The other gap runs the opposite way. Anthropic documents that auto-reload watches only the purchased balance and can charge your card while free credit remains. OpenAI says Auto Recharge fires “when your balance falls below the threshold you set” and doesn’t say which balance. Until it does, assume it can behave like Anthropic’s.

What should a developer paying for both ChatGPT Pro and Claude Max do with the API credits?

Claim both into fresh organizations with recharge off, check each one’s rate limits, and route work by what each credit is documented to cover. Self-run claude -p and Agent SDK jobs go on the Anthropic organization. Agents that lean on hosted tools, sandboxes or voice fit the OpenAI one better, because only OpenAI states those charges are covered. Codex and interactive Claude Code stay on the subscriptions.

Willison also says he may ship public-facing demos now that the credit covers them. Both labs’ credit terms forbid transferring or selling credits, though neither says anything against a public app those credits pay for. As a budget cap, a credit-only organization leaks a little. OpenAI’s billing page warns that access “may not stop immediately” and says outright:

Do not rely on the prepaid balance as an instantaneous spending cutoff.

Whatever runs over shows up as a negative balance, taken out of your next purchase. So give a demo its own OpenAI project or Claude workspace with a spend limit set and its own key, and handle each provider’s error when the money runs out: on OpenAI that is HTTP 429 with credit_balance_exhausted.