OpenAI cuts its API usage tiers from five to three, and the top one now costs $500
OpenAI has replaced the five numbered usage tiers on its API with three named ones, and the spend that unlocks the highest tier has halved. The company's API changelog entry for October 6 says it “Simplified API usage tiers from five to three: Build, Launch, and Grow. Organizations automatically upgrade as total credit purchases reach tier minimums.”

OpenAI’s own diagram of an application, the Agents API and a sandbox. Usage tiers set how fast any of it is allowed to run. Image: OpenAI
The table on OpenAI’s rate limits page, read at 22:05 UTC on October 7, lists a free tier capped at $100 a month for users “in an allowed geography”, then Build at $5 in total credit purchases for $500 a month, Launch at $100 for $5,000 a month, and Grow at $500 for $200,000 a month.
The old ladder charged twice as much at the top
Working out what moved means putting that live page next to a dated capture of the same URL. The Internet Archive’s capture of October 1 at 23:27 UTC still shows the old table, so what follows is what the page used to say and is no longer live.
In that version, Tier 1 needed $5 paid and allowed $100 a month, Tier 2 needed $50 for $500, Tier 3 needed $100 for $1,000, Tier 4 needed $250 for $5,000, and Tier 5 needed $1,000 for $200,000. The free row was unchanged at $100 a month.
Line the two up and the entry point stays at $5 while its ceiling rises fivefold, from $100 a month to $500. The $5,000 ceiling that used to need $250 now needs $100. The $200,000 ceiling that used to need $1,000 now needs $500. The wording of the qualification changed as well: the old table counted an amount “paid”, and the new one counts “total credit purchases”.
Dating the switch is easy enough from the archive. The October 1 capture carries the old five-row table, and a capture taken on October 6 at 22:13 UTC carries the three-row one, which brackets the change to the day the changelog gives it.
Rate limits now come in three steps, not five
The same page sets standard limits per tier. Build allows 5,000 requests a minute and 1,000,000 tokens a minute on Astra, Sol and Terra, or 2,000,000 on Luna. Launch doubles requests to 10,000 and lifts tokens to 4,000,000, or 10,000,000 on Luna. Grow runs at 15,000 requests and 40,000,000 tokens a minute, and 30,000 requests and 180,000,000 tokens on Luna.
Those are the standard limits. OpenAI’s page notes they differ from the separate Ultrafast rate limits, and it points organisations at Settings, Organization and then Limits in the platform console to see the figure attached to each model. The free row is the only one that is not bought: it qualifies on geography alone, through OpenAI’s list of supported countries, and it has kept its $100 monthly cap throughout.
Two different things on OpenAI’s books now carry a $500 label. The threshold above is a cumulative spend on API credit by a developer organisation, and it is not the $500 ChatGPT Pro plan that arrived at the end of September, which is a consumer subscription billed every month. The word tier is doing double duty as well: the Ultrafast processing tier on GPT-6 Astra is a per-request setting, not an account status, and it costs $300 per million output tokens rather than $500 of anything.
Nobody has to apply for any of this. OpenAI’s page says an organisation’s tier “upgrades automatically as its total credit purchases reach each threshold”, which means the first sign of a move is a higher ceiling rather than an email.