Five Hundred Dollars a Month: OpenAI's DevDay Pricing Earthquake Halves Pro Limits and Crowns a New Top Tier

It has become a running joke that every OpenAI developer conference ends with a pricing surprise. This year nobody is laughing — or perhaps only the people paying are not. Within minutes of Sam Altman’s DevDay 2026 keynote at Fort Mason in San Francisco, OpenAI confirmed a sweeping restructure of its consumer subscription ladder: a new Pro 500 plan at $500 per month with the company’s highest usage limits, a halving of what the existing $200 Pro tier buys you, the reopening of that same $200 tier after a weeks-long freeze, and a fresh user milestone — 1.2 billion weekly ChatGPT users — that explains why the company is squeezing supply so hard.

Taken together, the four moves read less like a product update and more like a utility renegotiating its rates after discovering that demand has outrun every forecast it ever made.

What actually changed

The new Pro 500 tier. The headline number is the price: five hundred dollars a month, two and a half times the cost of the existing Pro plan. In exchange, subscribers get OpenAI’s highest usage ceilings and — crucially — access to the new Ultrafast mode, which lets GPT-6 Astra deliver markedly faster responses across ChatGPT Work and Codex, up to eight times faster in the Codex case according to CNET’s recap of the keynote. Ultrafast currently runs on Astra, with GPT-6.1 Sol support promised soon. The subtext is hard to miss: raw speed is now a premium good, priced and rationed like one.

The $200 Pro tier got poorer. Existing Pro subscribers previously enjoyed roughly twenty times the usage allowance of a $20 Plus plan. That multiplier is being cut to ten — same price, half the goods. OpenAI is softening the blow with a one-time transition credit, though the company has not detailed its size, and current limits persist during a migration window whose end date is undisclosed. CNET’s framing was blunt: a clear sign that OpenAI is “trying to make as much money as possible from its subscribers while it burns through billions of investor dollars.”

The $200 tier reopened. In early September, demand for GPT-6 Astra strained OpenAI’s infrastructure so severely that the company suspended new Pro sign-ups entirely. At DevDay the plan came back, but with a rewritten usage calculation. Thibault Sottiaux of OpenAI wrote on X: “If you do the math, it will net out at half the dollar in API spend compared to the old Pro $200 plan.” A reopened door with a narrower frame.

1.2 billion weekly users. Jay Peters at The Verge notes the figure is up from the 1-billion milestone ChatGPT passed in July — meaning the service added roughly two hundred million weekly actives in about two months, a growth velocity that makes the infrastructure strain look less like an accident and more like an inevitability.

Why OpenAI is doing this now

The economic logic is not mysterious. Astra-class models are the most compute-hungry systems OpenAI has ever served, and a flat $200 plan purchased in 2024’s compute regime is ruinous in 2026’s. Every heavy Pro user streaming Astra at 20x Plus multiples was, in effect, being subsidized at a loss. Halving the multiplier and erecting a $500 ceiling above it converts an unbounded liability into a tiered ladder that monetizes exactly the users who cost the most.

The timing also serves the IPO narrative. OpenAI’s prospectus leaked last week revealed a $42 billion revenue ambition, and subscription restructures like this one are the most direct lever the company controls. It is one thing to promise growth to investors; it is another to demonstrate, in a single keynote, that your most engaged users will accept a 2.5x price increase for your best product.

There is also a competitive frame. Meta’s Muse agent topped App Store charts with 600,000 daily active users in the US earlier this month, and Google’s Gemini Spark fields a 24/7 consumer agent with more than thirty service partners. OpenAI’s answer — the Dots agents also announced at DevDay — needs an economic foundation that can absorb agentic workloads running continuously in the background. Ultrafast mode bound to a $500 tier is that foundation’s load-bearing wall.

What it means for users

For Plus subscribers, nothing changes today; the squeeze is aimed upward. For $200 Pro users, the math is unwelcome but survivable: unchanged price, halved relative allowance, and a transition credit of unknown generosity. The real decision falls on power users — the coders and agentic-workflow builders who live in Codex and ChatGPT Work. For them, Pro 500 is less a luxury than a new line item: if your monthly Astra consumption already brushes the new caps, $500 buys the headroom and the speed that the old plan used to provide.

The precedent, though, reaches further than any single tier. OpenAI has now demonstrated that usage multipliers can be cut retroactively — that the deal you subscribed to is negotiable when compute economics shift. That is a rational move for a company rationing constrained silicon, and a cautionary one for an industry racing to build always-on agents on top of subscriptions written for a lighter era. When the next demand shock arrives, every lab now has a template: reopen the plan, rewrite the math, add a pricier rung above it.

The bigger picture

DevDay 2026 will be remembered for Dots and for GPT-6.1 Sol’s budget-friendly positioning, but the subscription grid is the announcement with the longest half-life. It formalizes a new physics for consumer AI pricing: intelligence is metered, velocity is a premium feature, and a billion-plus weekly users is both OpenAI’s greatest asset and its most expensive liability. The company that popularized the flat-rate chatbot subscription has, one tier at a time, taught the market that compute is a utility — and utilities, as anyone who has ever opened a power bill knows, do not stay cheap forever.

For developers and enterprises, the actionable takeaway is to audit your usage now. The transition credit is finite, the migration window is vague, and the price of speed has a comma in it. The era of AI as an all-you-can-eat buffet is ending; DevDay 2026 is the moment it acquired a menu.