OpenAI Doubled ChatGPT's Usage Limits After a Revolt — and Its $500 Tier Now Gives Back $500 in Free API Credits

The Sept. 29 package was textbook ladder-building: a new top rung to pull power users upward, and a 50% cut to the allowance of the $200 tier to make the jump feel necessary. The backlash was immediate, and the fix was surgical — the allowance, not the price, is what OpenAI gave back.

The details of what was restored matter as much as the fact of it. Existing Pro 200 subscribers kept the higher allowance with no waiting for renewal, the weekly GPT-6 Pro message limit on Pro 200 was raised to match the higher tier, and Business Premium was doubled along with the Pro tiers, which tells you the reversal was about the underlying Work and Codex capacity pool rather than a single legacy plan.

Two gaps remain, and they are the ones an investor cannot fill from outside. OpenAI has not published the numeric allowance multiple behind Pro 100 or Pro 500, and it confirmed that extended usage limits for its Dots product continue separately from default usage while declining to publish per-tier figures. The shape of the restoration is known; the size of the giveaway is not.

OpenAI's own forecast, reported in April 2026, had ChatGPT Plus subscribers falling 80% from 44 million in 2025 to about 9 million in 2026, with the $8 ad-supported Go tier growing roughly 36 times, from 3 million to 112 million, and Pro users expected to double but stay below 1% of the total. On those numbers a $500 rung cannot move the revenue mix on its own, and the marginal consumer dollar is being harvested at the bottom of the ladder and in advertising: ChatGPT Ads passed a $1 billion annualised run rate in under 200 days, with the free tier supporting more than 1 billion weekly active users.

The backlash, in other words, was about what users received, not what they paid. That is why the company restored quantity across every paid tier, including its newest one, and changed no price at all: a discount would have tested the wrong variable.

The decisive detail landed on Oct. 28, one day before the reversal: eligible ChatGPT Pro and Business subscribers can claim free API credits sized to their subscription — $100 a month on Pro 100, $200 on Pro 200, $500 on Pro 500, $20 and $100 per Business seat pooled per workspace up to $500 a month, and up to $6,000 a year for workspaces billed annually. The credit equals the subscription price at every tier. A Pro 500 subscriber therefore pays $500 and receives $500 of metered API usage, with the doubled Work and Codex allowance on top of it.

That is not a premium tier; it sells inference in advance at OpenAI's own internal cost, with the rebate standing in for the compute already committed to Stargate. The economics flip: subscription cash becomes a fixed, contracted demand signal for capacity that is being built years ahead, which is exactly the mix that protects enterprise and API demand while consumer ARPU stays small.

Follow the money and the ladder looks less like a consumer business than a demand-acquisition funnel for compute. Microsoft recognised $10 billion of revenue from OpenAI commercial arrangements, including revenue-share payments, in the quarter ended Sept. 30, 2026 — nearly 11% of its $92.3 billion of revenue — after $24.1 billion in fiscal 2026, with the share due to continue through 2030 subject to a cap reported at $38 billion.

Anthropic prices the same frontier from the other side of the ladder. Its Oct. 26 S-1 shows $16.29 billion of first-half revenue, up from $1.16 billion a year earlier, of which $13.1 billion — 80% — was consumption and only $3.2 billion was subscription, at roughly 85% gross margin, with net revenue retention near 900% and a first-half net loss of $80.7 billion driven largely by a one-time charge. Claude's consumer ladder stops at $200 for Max 20x, so OpenAI's $500 rung sits above anything Anthropic sells to individuals.

And xAI, now inside SpaceX, chose a third route: bundling Grok, Cursor and X Premium from $8 to $200 a month, with team seats priced identically to individual seats. SpaceX's Q3 was the first with a profitable AI segment — Super Intelligence revenue of $7.56 billion, up 830% year over year, $431 million of segment operating income, and 1.7 gigawatts of nameplate compute.

Every extra token of ChatGPT Work and Codex usage, and every $500 of rebated API credit, has to be served somewhere, and OpenAI has been buying that capacity in advance: nearly 7 gigawatts of planned Stargate capacity and more than $400 billion of investment over three years, including an OpenAI-Oracle partnership exceeding $300 billion over five years with up to 4.5 additional gigawatts.

That is the part investors can actually underwrite. A doubled allowance turns a subscription price into a capacity order, and the gross margin table below shows where the economics of that order are retained.

Near term, three dated checkpoints matter more than the tier ladder itself: Anthropic's IPO pricing after the Oct. 26 S-1, Microsoft's FY27 Q2 and Q3 disclosures where OpenAI-related revenue can be tracked against the $38 billion cap, and CoreWeave's Nov. 10 third-quarter call, where consumer-driven demand shows up first because its contracts are short and its balance sheet is levered.

Over one to three years the test is simple: whether a sub-1% user segment can keep funding gigawatts of new capacity. If it can, the rebate on Pro 500 is a rational way to reserve capacity; if consumer-tier ARPU stalls, the same promise becomes a margin transfer to whoever supplies the tokens.