OpenAI's $500 Pro 500 Plan Arrives as $200 Pro Limits Get Halved

OpenAI added a new top tier to ChatGPT at DevDay 2026: Pro 500, a $500-per-month subscription that includes the new Astra Ultrafast speed tier and a usage allowance the company describes as roughly 25 times Plus. The louder story may be the quieter one beneath it: for new subscribers, the existing $200 Pro plan is being reduced — ChatGPT Work and Codex usage falls from 20x to 10x Plus equivalency, and weekly GPT-6 Pro messages drop from 200 to 100. Existing Pro subscribers keep their current limits temporarily and receive one-time credits.

This piece lays out exactly what changed across the subscription ladder, what Pro 500 buys that Pro no longer does, and how the arithmetic works for heavy users deciding between tiers or shifting workloads to the API. It is written for individual power users and team leads who own the ChatGPT line item.

The new ladder: what each tier now gets

Pro 500 sits above everything: $500 per month, around 25x Plus usage, and access to Astra Ultrafast — GPT-6 Astra served at up to 8x speed in Codex, roughly 300 tokens per second — inside ChatGPT Work and Codex, an option otherwise reserved for Enterprise. The $200 Pro plan relaunched for new subscribers the same day with its reduced allowances: 10x Plus for Work and Codex, and 100 GPT-6 Pro messages per week.

The old Pro terms do not vanish overnight for people already paying — OpenAI says existing subscribers maintain current limits temporarily, cushioned with one-time credits — but the direction is explicit. What was one $200 tier with 20x usage is now two tiers, and the capability that used to define Pro has been split: half the usage stays at $200, while the full allowance, plus Ultrafast, costs $500.

Why OpenAI restructured now

DevDay 2026's launches all consume compute aggressively: dots work around the clock on their own cloud computers, Codex now runs entire tasks in the cloud, and Ultrafast dedicates serving capacity to speed. Pricing the heaviest interactive users into a $500 tier is the straightforward way to fund that — and the announcement pattern matches, with Ultrafast gated to exactly Enterprise and Pro 500. Notably, dots conversations do not count against usage limits, but Codex and Work tasks a dot starts do, which routes always-on agent activity back into the metered allowances this restructure just repriced.

There is also a segmentation logic: GPT-6.1 Sol, launched the same day at one-fifth of Astra's API price, gives cost-sensitive users a cheaper road, while Pro 500 gives latency-sensitive users a faster one. The $200 tier's halving pushes anyone in between to pick a direction.

Who is affected, and the arithmetic that matters

New subscribers bear the full change immediately: $200 now buys half the Work and Codex usage it bought last week, and half the GPT-6 Pro messages. For current Pro subscribers the question is timing — 'temporarily' is not a date, and the one-time credits soften rather than prevent the transition. Anyone whose workflows assume 20x usage should measure actual consumption now, before the old limits lapse.

The comparison worth running: a Pro user consistently hitting the new 10x ceiling is paying for usage that Pro 500 prices at $500 for 25x — but the API prices the same models with no subscription at all, and GPT-6.1 Sol matches near-Astra quality at $2 per million input tokens. For programmatic and agent workloads, the API route got dramatically cheaper the same day the subscription route got tighter; the subscriptions increasingly sell interactivity and Ultrafast speed, not bulk tokens.

The competitive context

A $500 consumer-adjacent AI subscription resets the market's reference point for what individual power users might pay — the previous ceilings across the industry sat at the $200 mark that Pro established. Whether rivals follow depends on whether Ultrafast-style speed tiers become table stakes; OpenAI has effectively unbundled speed as a separately billable dimension, in the API at a 6x multiplier and in subscriptions via Pro 500.

The reduction pattern is worth watching on its own. Subscription allowances that quietly shrink for new cohorts while grandfathering existing ones are a familiar SaaS maneuver, but applying it to the plan level right as always-on agents increase background consumption changes effective per-dollar value in ways headline pricing does not show.

Grandfathering, credits and caps: what to watch

Three specifics are unresolved: how long existing Pro subscribers keep 20x limits, what the one-time credits amount to in practice, and whether the 100-message GPT-6 Pro weekly cap changes as GPT-6.1 Sol absorbs routine workloads. Watch also for GPT-6.1 Sol Ultrafast, due within days — if it lands inside Pro rather than Pro 500, the middle tier regains some of its lost appeal.

The practical step this week: pull your actual usage against the new definitions before renewing anything. If your consumption fits inside 10x Plus, the reduced Pro changes nothing for you; if it does not, price the delta three ways — Pro 500, Enterprise, or moving the heavy workloads to gpt-6.1-sol on the API — because after DevDay 2026 those are genuinely different products.