Announcement details
On September 29, at DevDay 2026 in San Francisco, OpenAI announced a new top-tier ChatGPT subscription — Pro 500. The new tier costs US$500 per month — 2.5x the previous most expensive tier, Pro 200 ($200/month).
According to the company, Pro 500 subscribers get a usage allowance 25x the ChatGPT Plus limit. The figure targets compute-heavy tasks first: working with large codebases, analyzing hundred-page documents, and running multi-step agentic jobs.
The ChatGPT Pro lineup now spans three tiers: $100, $200 and $500. With this move, OpenAI is cleanly segmenting professional users by workload intensity.
Astra Ultrafast — only on the priciest tier
One of DevDay's most notable technical announcements was the Astra Ultrafast model. Per OpenAI, the model runs 8x faster than usual in the Codex environment, reaching 300 tokens per second; via the API, a 6x speedup was reported.
Crucially, access to Astra Ultrafast is granted only to Pro 500 subscribers and select Enterprise and Edu customers. The company's fastest model sits behind the priciest subscription wall — a clear expression of a "pay separately for speed" model.
But speed has a hidden price: the company warns that running in Ultrafast mode burns through the usage allowance 8x faster. So the advertised "25x" limit will in practice run out much sooner under Ultrafast — an important detail when choosing a tier.
Per OpenAI's official announcement, Astra Ultrafast runs 8x faster in Codex (300 tokens/sec) and 6x faster via the API — but that speed is open only to Pro 500 subscribers and select enterprise customers, and the allowance burns 8x faster.
Pro 200: reopened, but limits cut
In an interesting twist, OpenAI simultaneously reopened the Pro 200 tier to new subscribers — previously it had been kept only for existing users. The reopening came with a significant cut to limits: the multiplier vs Plus dropped from 20x to 10x, and the weekly message limit for GPT-6 Pro was cut from 200 to 100.
A transition period was set for existing Pro 200 subscribers: they keep the old limits until October 29, 2026, and receive a one-time credit on top. That's the standard way to soften sharp pricing-policy changes for existing customers — but it is time-limited too.
What this move means
The Pro 500 launch together with the Pro 200 limit cuts paints one picture: access to the most powerful AI models and high-volume usage is increasingly becoming a precisely priced, strictly segmented product. OpenAI is moving from the era of "generous limits" to an era of calculated, predictable monetization.
Competitively, the move piles pressure on professional tiers from Anthropic, Google and xAI. The contest is shifting from "who shipped the stronger model" to "who offers heavy users the best price/performance ratio" — and for enterprise customers that will be the deciding factor. For users, though, the choice gets harder: without understanding each tier's real boundaries, making the right call becomes difficult.
The trend also shows AI compute turning into a "utility": like electricity or cloud billing, AI compute is becoming a metered, tariffed, budgetable resource.
Uzbekistan context
In Uzbekistan, $500/month is a serious sum for most freelancers and small teams. But the decision shouldn't be individual — it should rest on team-level math: if a team spends several hours a day on heavy code generation or large-document analysis, Pro 500's 25x allowance may work out cheaper than the equivalent volume paid via API — worth calculating precisely.
Two practical notes. First, international payments from Uzbekistan face card and currency restrictions; for a corporate subscription, applying officially via the Enterprise channel is recommended. Second, the "25x" figure can mislead if you ignore Ultrafast burning the allowance 8x faster: it's wiser to measure the team's real workload for a week or two first, then choose a tier.
