ChatGPT is also launching a Pro Max version.
Rumors circulating today claim that OpenAI is preparing ChatGPT Pro Max, a subscription tier priced at up to $500 per month—2.5 times the cost of its current top-tier plan—with a single key selling point: “The fastest Work and Codex.”
It is estimated that the $600 listed here includes a 20% "channel tax."
Two weeks ago, it paused new subscriptions to the $200 Pro tier due to the GPT-6 Astra cluster overload;
Behind Astra, a stronger model rumored to be codenamed Bel, which has allegedly provided a proof for the Millennium Prize problem known as Navier–Stokes, is waiting in line to enter.
Computing power is in short supply; top-tier specs have doubled in price. As the most advanced productivity tools become increasingly expensive, a pressing question emerges: Will the next truly groundbreaking intelligence become a privilege reserved only for the highest bidder?
What did you buy for $500?
On September 24, developer Tibor Blaho and the leak site TestingCatalog independently discovered traces of Pro Max in ChatGPT’s unpublished frontend configuration: a monthly fee of $500, with key features labeled as “the fastest Work and Codex.”
https://x.com/btibor91/status/2103222629171908937
https://x.com/testingcatalog/status/2103238785278312463
If ultimately launched, it will directly surpass Claude Max ($250) and Google AI Ultra (~$250), setting a new ceiling for mainstream personal AI subscriptions.
Unlike before, this time what’s being purchased at a premium is not usage, but computing power priority.
Previously, users exchanging from the $100 tier to the $200 tier received 5 to 20 times the call quota of Plus;
The Pro Max targets time cost—complex tasks like those in Work and Codex can run continuously for hours, and for heavy productivity scenarios, low latency directly translates to profit.
There is speculation that its underlying infrastructure may be integrated with Cerebras low-latency hardware currently under testing by OpenAI.
Astra drained the hash power pool.
The direct trigger for the price increase was the exhaustion of mining power.
GPT-6 Astra, launched on September 3, features agent capabilities for direct system control and consumes significantly more reasoning compute power than its predecessor.
On September 10, just one week after launch, OpenAI urgently halted new subscriptions to the $200 Pro tier.
Tibo, who oversees the ChatGPT and Codex services, publicly stated that this tier "placed the greatest strain on the system," and OpenAI has not yet provided a timeline for restoration.
The business logic is not hard to understand.
The $200 tier has become a "heavy-duty all-you-can-eat buffet" for tech enthusiasts and professional institutions, causing the platform to bear substantial subsidy losses.
Rather than letting its system be overwhelmed by this extreme usage, OpenAI might as well embrace it by creating a separate tier with pricing closer to the actual inference cost.
The world's top intelligence is now openly priced.
Even greater capabilities are yet to come—GPT-6 Astra is just an appetizer.
The official previously revealed that the comprehensive performance of its latest internal model is "significantly superior to Astra," having cracked the proof of the Navier–Stokes equations.
The industry's pathways for allocating cutting-edge capabilities are diverging: OpenAI tends to tier access by commercial pricing, while Anthropic increasingly limits access based on safety risks and identity verification.
The latest update to Claude's official documentation states that Opus 5.5 will automatically downgrade high-risk tasks involving offensive cybersecurity, dual-use biology, and similar areas to a lightweight model. Only users who pass strict identity verification may apply for unblocking.
Additionally, Opus 5.5 will detect whether cutting-edge AI research is being conducted; if detected, it will reduce its intelligence and refuse to "work for others."
$500 per month (approximately RMB 3,352) almost draws a clear line: it is accessible only to businesses, institutions, or a very small number of individuals with high earning potential.
If future Bel-level breakthroughs are ultimately accessible only to those with $500 or more, technological fragmentation is likely to rapidly translate into real-world disparities—those with ample capital will leverage top-tier reasoning to snowball their advantages, while ordinary users remain below the mainstream baseline.
But this is also a cold business reality.
OpenAI is not a nonprofit organization; depreciation of cutting-edge hardware and electricity are direct costs.
In the same week, OpenAI also halved the API pricing for the lightweight model GPT-6 Sol and Luna.
General-purpose computing continues to rapidly depreciate due to deflation, while the only thing experiencing true inflationary surge is the extremely scarce frontier computing power at the very top of the pyramid.
There are still uncertainties surrounding DevDay next week.
Next Tuesday’s DevDay in the U.S. could be the earliest possible moment for Pro Max to be unveiled, alongside rumored new pricing tiers for the OpenAI Platform starting at $50.
At that time, there will be three key points for external verification: whether the $500 pricing will ultimately be implemented, when the $200 tier will be unfrozen, and at what cost the true next-generation foundation, Bel, will be unveiled.
After the first DevDay in 2023, OpenAI had to temporarily shut down access to its $20 Plus subscription due to an unexpected surge in traffic;
Three years have passed, and while the new DevDay hasn't opened yet, the $200 door has already closed.
Behind closed doors, a new $500 threshold has quietly been written into the code.
Reference materials:
https://x.com/testingcatalog/status/2103238785278312463
https://x.com/btibor91/status/2103222629171908937
https://support.claude.com/en/articles/16049681-why-claude-switched-models-in-your-conversation-with-opus-5-or-opus-5-5
Edited by Marco
This article is from the WeChat public account "New Intelligence Yuan" (ID: AI_era), authored by ASI Revelation.