CONFIRMED: OpenAI launches ChatGPT Pro 500 with speed powered by Cerebras chips
OpenAI confirmed at its September 29 DevDay the plan that had leaked days earlier: ChatGPT Pro 500, a 500 dollar monthly subscription, two and a half times above the 200 dollar Pro tier. The differentiator is not a smarter model, it is pure speed. The new plan is the only one that includes Ultrafast, a mode that speeds up GPT-6 Astra by up to eight times inside Codex and six times through the API, reaching 300 tokens per second. The live demo stumbled a few times on stage, which did not stop OpenAI from going ahead with the launch.
What changes in practice for people paying more
Pro 500 subscribers get twenty five times the usage included in Plus, access to Ultrafast in both ChatGPT Work and Codex, and priority queueing during peak hours. Ultrafast is already available for GPT-6 Astra and is coming soon for GPT-6.1 Sol, the cheaper model launched the same week for everyday tasks. On the API, the speed boost costs six times the standard rate: for Astra that means 60 dollars per million input tokens and 300 dollars per million output tokens, versus 10 and 50 dollars in standard mode. The rumor published before the event pointed exactly to this combination of high price and gains concentrated in speed, and the official announcement confirmed that reading point by point.
Cerebras officially enters the picture
The missing piece from the leak was also confirmed: Ultrafast runs on Cerebras chips, the dinner-plate sized processors known as wafer-scale, which OpenAI has used since an August preview. Cerebras markets its architecture as far faster than the current generation of Nvidia GPUs at a lower cost per chip, but the 500 dollar monthly price for access still looks steep against the efficiency gains the chipmaker publicly advertises. OpenAI has not publicly detailed how many Cerebras chips power the service nor the margin it charges on top of inference cost, which leaves the price versus efficiency comparison as a math only the company itself can fully close.
The other side of the coin: the $200 Pro tier gets thinner
While the top tier bulks up, the middle tier loses weight: starting October 30, the $200 Pro plan will include ten times Plus usage in ChatGPT Work and Codex, half of what it offered before, and weekly chat messages with GPT-6 Pro drop from 200 to 100. Current subscribers keep their old limits until October 29 and get a one time credit of 2,500 dollars in usage, valid through year end. In practice, OpenAI is pushing heavy Work or Codex users toward the pricier plan while reopening Pro to new subscribers with a smaller package than the original.
MaxAssistant's read
The launch confirms what the earlier leak already pointed to, including the near-exact name and the precise monthly price, proof that the convergence of technical sources ahead of the official announcement worked as genuine anticipation, not loose guessing. The thesis behind Pro 500 is to split intelligence from speed as two distinct products inside the same subscription, using third party hardware to justify the higher price. That is consistent with an OpenAI that, in the same week, launched GPT-6.1 Sol for affordability and Astra for capability, and now Ultrafast for those who cannot wait for the response. For automation, support, and product operations running heavy agents in Codex or Work, it is worth measuring whether the real throughput gain justifies the jump from 200 to 500 dollars before migrating, because Ultrafast charges by token volume processed, not just by the subscription fee.
Sources
OpenAI: DevDay 2026 Recap; The Decoder: OpenAI expands Codex and its API at DevDay with Ultrafast; The Verge: OpenAI is adding a $500/month ChatGPT Pro tier.