OpenAI Cuts ChatGPT Limits, Launches $500 Pro Plan
OpenAI reduced ChatGPT usage caps and introduced a $500 Pro tier. Prices remain static, but the value received by subscribers has decreased.
English · Original discussion in Spanish · Published
OpenAI Cuts ChatGPT Limits, Launches $500 Pro Plan
On September 29, 2026, OpenAI restructured its subscription catalog with the arrival of Pro 100, Pro 200, and Pro 500 plans, according to a participant in the discussion. On paper, the $200 subscription maintains its price. In practice, new subscribers at this level receive less included usage than existing ones, who retain their previous quota only until October 29, 2026, according to the same source. It is a disguised price increase per unit of use, framed as tariff continuity, they argue.
The complaint circulating through consumer forums is not the nominal price hike, but the 50% reduction across all tiers that users attribute to the service. Those paying $20 a month for the Plus plan claim they previously stretched the weekly limit over five or six days; now they exhaust it in two. And not with expensive models, but with those the company itself sells as economical.
What Exactly Changed in ChatGPT Plans
The move has three legs, according to information circulating in the thread. The first is the quota reduction in existing levels. The second is the creation of a higher tier, the Pro 500 at $500 monthly, presented as the only one with access to Astra Ultrafast, the maximum-estimulante ilegal mode. The third is the fine print: former Pro 200 subscribers temporarily keep the previous allowance, but new ones do not.
That is the detail worth highlighting. The company did not touch the price of the $200 subscription. It touched what that money buys, according to users reporting the issue. It is the same play used by telecoms with fiber optics and utilities with night rates: keeping the storefront while cutting the content. Gym psychology: the membership fee doesn't go up, but you can no longer use the pool.
The Calculation That Debunks the Cheap Plan Narrative
There is a recurring argument defending the $20 Plus plan: that for €23 a month—the approximate exchange rate cited—the difference with the free version is vast, and anyone haggling over that amount spends it on a set menu or two packs of cigarettes. The reasoning makes sense until one looks at actual usage.
The problem isn't the price. It's the usage density allowed by the limit. A developer working with the model for several hours a day exhausts the weekly quota in two days, according to testimonials in the thread. From there, they either pay more, wait, or migrate. The comparison with the daily special falls apart when the special runs out on Tuesday.
User Response: Local AI and Chinese Models
The dominant reaction isn't cancellation, it's redistribution. More and more people are setting up open-source models on their own computers and using them for repetitive tasks: formatting, summarizing, boilerplate coding. Paid cloud services are reserved for tasks requiring fine reasoning. It is a two-layer architecture that sounded like geekery a year ago and is now explained in any office conversation.
Hardware supports this. A 30B model quantized to 4 bits fits on a 24 GB GPU; a 14B model fits on 12-16 GB, according to data discussed in the thread. For larger volumes, one must distribute across multiple cards or accept that tokens per second plummet. The bottleneck isn't the processor, but memory bandwidth: hundreds of GB/s on a GPU versus tens on system RAM.
The other leg is Chinese. Open-source models released by Asian companies don't match proprietary frontier models, but the gap narrows every quarter, according to participants. The strategy is transparent: give away market share while American companies raise prices. For the average user, the practical consequence is that DeepSeek or Qwen cover 80% of daily tasks without paying, according to a calculation circulating in the debate.
The Energy Cost No One Includes in the Equation
There is a factor appearing in the conversation that usually stays out of subscription analyses: energy. Training and inferring frontier models consumes electricity at a rate that strains grids, as noted in the thread. The energy crisis isn't just background noise; it's a cost variable that will eventually be passed on to the final price. When that happens, today's $200 will seem like a bargain.
OpenAI's silent price hike isn't an isolated case. It is the first visible move by a sector that spent three years burning cash to build dependency and is now starting to charge for it, according to users. The question isn't whether it will rise. The question is how long it will take others to copy the playbook.
How long can a company sustain the narrative that prices haven't changed when what has changed is what you get for them?
On September 29, 2026, OpenAI restructured its subscription catalog with the arrival of Pro 100, Pro 200, and Pro 500 plans, according to a participant in the discussion. On paper, the $200 subscription maintains its price. In practice, new subscribers at this level receive less included usage than existing ones, who retain their previous quota only until October 29, 2026, according to the same source. It is a disguised price increase per unit of use, framed as tariff continuity, they argue.
The complaint circulating through consumer forums is not the nominal price hike, but the 50% reduction across all tiers that users attribute to the service. Those paying $20 a month for the Plus plan claim they previously stretched the weekly limit over five or six days; now they exhaust it in two. And not with expensive models, but with those the company itself sells as economical.
What Exactly Changed in ChatGPT Plans
The move has three legs, according to information circulating in the thread. The first is the quota reduction in existing levels. The second is the creation of a higher tier, the Pro 500 at $500 monthly, presented as the only one with access to Astra Ultrafast, the maximum-estimulante ilegal mode. The third is the fine print: former Pro 200 subscribers temporarily keep the previous allowance, but new ones do not.
That is the detail worth highlighting. The company did not touch the price of the $200 subscription. It touched what that money buys, according to users reporting the issue. It is the same play used by telecoms with fiber optics and utilities with night rates: keeping the storefront while cutting the content. Gym psychology: the membership fee doesn't go up, but you can no longer use the pool.
The Calculation That Debunks the Cheap Plan Narrative
There is a recurring argument defending the $20 Plus plan: that for €23 a month—the approximate exchange rate cited—the difference with the free version is vast, and anyone haggling over that amount spends it on a set menu or two packs of cigarettes. The reasoning makes sense until one looks at actual usage.
The problem isn't the price. It's the usage density allowed by the limit. A developer working with the model for several hours a day exhausts the weekly quota in two days, according to testimonials in the thread. From there, they either pay more, wait, or migrate. The comparison with the daily special falls apart when the special runs out on Tuesday.
User Response: Local AI and Chinese Models
The dominant reaction isn't cancellation, it's redistribution. More and more people are setting up open-source models on their own computers and using them for repetitive tasks: formatting, summarizing, boilerplate coding. Paid cloud services are reserved for tasks requiring fine reasoning. It is a two-layer architecture that sounded like geekery a year ago and is now explained in any office conversation.
Hardware supports this. A 30B model quantized to 4 bits fits on a 24 GB GPU; a 14B model fits on 12-16 GB, according to data discussed in the thread. For larger volumes, one must distribute across multiple cards or accept that tokens per second plummet. The bottleneck isn't the processor, but memory bandwidth: hundreds of GB/s on a GPU versus tens on system RAM.
The other leg is Chinese. Open-source models released by Asian companies don't match proprietary frontier models, but the gap narrows every quarter, according to participants. The strategy is transparent: give away market share while American companies raise prices. For the average user, the practical consequence is that DeepSeek or Qwen cover 80% of daily tasks without paying, according to a calculation circulating in the debate.
The Energy Cost No One Includes in the Equation
There is a factor appearing in the conversation that usually stays out of subscription analyses: energy. Training and inferring frontier models consumes electricity at a rate that strains grids, as noted in the thread. The energy crisis isn't just background noise; it's a cost variable that will eventually be passed on to the final price. When that happens, today's $200 will seem like a bargain.
OpenAI's silent price hike isn't an isolated case. It is the first visible move by a sector that spent three years burning cash to build dependency and is now starting to charge for it, according to users. The question isn't whether it will rise. The question is how long it will take others to copy the playbook.
How long can a company sustain the narrative that prices haven't changed when what has changed is what you get for them?
Also available in: Deutsch
Summary of a discussion on Burbuja.info - Foro de economía, actualidad y política., translated from Spanish and reviewed before publication. Read the full discussion (83 replies).
More summaries
-
Albo and frozen Louriño, the most recommended callos on the forum
Several forum members criticise Litoral and recommend Albo, La Nora and, above all, the frozen Louriño
-
Driving License No Longer a Rite of Passage for Adults
Many adults live without a driving license as car costs, free public transport, and fear outweigh its symbolic value.
-
Foster's Hollywood in Spain: Shrinking Portions, Rising Prices
Returning customers to Foster's Hollywood in Spain report shrunken portions, industrial bread on €14 burgers, and higher prices.
-
Spanish 1970s Cookies: The Lost Taste of Childhood
Iconic Spanish brands like Cuétara and Fontaneda have changed. Palm oil, mergers, and quality cuts explain the decline of a national staple.
-
Bluetooth Headphones: The Comfort Scam Exposed
Are Bluetooth headphones worth it? We analyze price, sound quality, battery life, and health risks. Data suggests wired options still win.
-
Andreu Bacardit Promises Free Energy, Raising Fraud Concerns
The Onion Project claims infinite clean energy, but its promoter has a history of failed products. We analyze the red flags and potential scam patterns.