Enshittification reaches AI: the free tier is switched off
Google shuts down free access to Gemini Pro and Flash, video generation is being wound down and paid models now cost up to €500 a month. Enshittification is slowing the AI revolution.
English · Original discussion in Spanish · Published
Enshittification reaches AI and switches off free access
How long will it take for artificial intelligence to stop being free? The answer, judging by the state of the sector as of this discussion, is that it is already happening. Google has reportedly announced the closure of its Gemini Pro and Flash models for free use, leaving only Flash Lite, described as unusable for almost any task. Video generation, which a year ago offered a thousand options at no cost, has been left in beta with low-quality models. And image generation, with tools like Nanobanana 2, is heading down the same path.
The pattern has a name in English: enshittification. The term describes the deliberate degradation of a platform to push users toward a paid version worse than the original. The classic example cited is the jump from Windows 7 to Windows 8. The new one is generative AI.
What has peine with Gemini, Grok, ChatGPT and Claude
The sequence is familiar. First, free and massive access to train the models with real usage data. Then, once the product matures, the paywall goes up. Google, according to the premise that opens this discussion, has reportedly closed Gemini Pro and Flash to free users, joining similar moves at Grok, ChatGPT and Claude. The result: the barrier to entry now sits at around 100 euros a month for an acceptable version, and the higher tiers reach 500 euros a month.
Some argue the business math is flawless: the models advanced thanks to mass usage, and now it is time to charge. Against that weighs an argument the industry itself should antiestéticar: if access becomes more expensive, the volume of interaction falls, and with it the quality of the data feeding the next generation of models. The revolution stalls just when it promised to accelerate.
Video generation: the case that anticipates what is coming
Generative video is the laboratory where the full cycle has already played out. The technology was American, with a big lead. It was put behind a paywall. And the Chinese models, with Seed Dance at the front, seized the opening. Google launched Omni, described as stunning, but charging for it cost it traction: those who already had a workflow built on Seed Dance did not move. The lesson is uncomfortable for American companies: charging too early hands over the market.
The Chinese case introduces another variable. In language models, Chinese developments lag behind, which allows the Americans to enshittify everything without losing ground. But that advantage is not eternal. If the gap closes, the business model based on degrading the free version becomes unsustainable.
The real cost of running your own AI at home
Against the paywall, the alternative of running your own server emerges. The technical answer is blunt: to run Kimi3, the only open-source model with quality comparable to Gemini 3.8 Flash, you need around 600 gigabytes of RAM. It is not a home solution for most people. The promise of free software collides with the reality of hardware.
The figure dismantles the easy argument of "set up your own server." For the average user, the real choice is to pay or be left out. And that, in a sector that feeds on mass usage, is a problem the companies themselves have created.
The addiction business: free candy, then the uphill climb
The metaphor that best describes the model is the drug dealer's. First free candy. Then, once you are hooked, they tell you where you have to go. It is the same pattern as the first drink of any substance: the house buys the first round. Applied to AI, the user who has integrated these tools into their work discovers that the version they used no longer exists.
The open question is whether this model is sustainable in the long run. Technology, as someone quoting a certain Ted recalled, never retreats: it only adheres. But adoption can stall. And without mass adoption, the promise of replacing workers and multiplying profits stays stuck in a PowerPoint.
In the end, the conclusion is the usual one: we could have abundance for everyone, but someone has decided it is better to charge for it monthly. And meanwhile, the Chinese are waiting.
How long will it take for artificial intelligence to stop being free? The answer, judging by the state of the sector as of this discussion, is that it is already happening. Google has reportedly announced the closure of its Gemini Pro and Flash models for free use, leaving only Flash Lite, described as unusable for almost any task. Video generation, which a year ago offered a thousand options at no cost, has been left in beta with low-quality models. And image generation, with tools like Nanobanana 2, is heading down the same path.
The pattern has a name in English: enshittification. The term describes the deliberate degradation of a platform to push users toward a paid version worse than the original. The classic example cited is the jump from Windows 7 to Windows 8. The new one is generative AI.
What has peine with Gemini, Grok, ChatGPT and Claude
The sequence is familiar. First, free and massive access to train the models with real usage data. Then, once the product matures, the paywall goes up. Google, according to the premise that opens this discussion, has reportedly closed Gemini Pro and Flash to free users, joining similar moves at Grok, ChatGPT and Claude. The result: the barrier to entry now sits at around 100 euros a month for an acceptable version, and the higher tiers reach 500 euros a month.
Some argue the business math is flawless: the models advanced thanks to mass usage, and now it is time to charge. Against that weighs an argument the industry itself should antiestéticar: if access becomes more expensive, the volume of interaction falls, and with it the quality of the data feeding the next generation of models. The revolution stalls just when it promised to accelerate.
Video generation: the case that anticipates what is coming
Generative video is the laboratory where the full cycle has already played out. The technology was American, with a big lead. It was put behind a paywall. And the Chinese models, with Seed Dance at the front, seized the opening. Google launched Omni, described as stunning, but charging for it cost it traction: those who already had a workflow built on Seed Dance did not move. The lesson is uncomfortable for American companies: charging too early hands over the market.
The Chinese case introduces another variable. In language models, Chinese developments lag behind, which allows the Americans to enshittify everything without losing ground. But that advantage is not eternal. If the gap closes, the business model based on degrading the free version becomes unsustainable.
The real cost of running your own AI at home
Against the paywall, the alternative of running your own server emerges. The technical answer is blunt: to run Kimi3, the only open-source model with quality comparable to Gemini 3.8 Flash, you need around 600 gigabytes of RAM. It is not a home solution for most people. The promise of free software collides with the reality of hardware.
The figure dismantles the easy argument of "set up your own server." For the average user, the real choice is to pay or be left out. And that, in a sector that feeds on mass usage, is a problem the companies themselves have created.
The addiction business: free candy, then the uphill climb
The metaphor that best describes the model is the drug dealer's. First free candy. Then, once you are hooked, they tell you where you have to go. It is the same pattern as the first drink of any substance: the house buys the first round. Applied to AI, the user who has integrated these tools into their work discovers that the version they used no longer exists.
The open question is whether this model is sustainable in the long run. Technology, as someone quoting a certain Ted recalled, never retreats: it only adheres. But adoption can stall. And without mass adoption, the promise of replacing workers and multiplying profits stays stuck in a PowerPoint.
In the end, the conclusion is the usual one: we could have abundance for everyone, but someone has decided it is better to charge for it monthly. And meanwhile, the Chinese are waiting.
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 (22 replies).
More summaries
-
Guess who started the feminist and Marxist movement in Spain
Margarita Nelken, born in Madrid in 1894 to a German-Jewish family, is closely linked to the origins of feminism and Marxism in Spain. Learn about her legacy.
-
Four 'Valencias' or twelve 'Valladolids' to build 2 million homes
Two AI models propose 4 megacities of 500,000 homes each or 12 cities of 150,000 to build 2 million houses and bring down rents in Spain.
-
Forum user claims over €1,200 in Spanish minimum income benefit for three
A forum user says he receives more than €1,200 under Spain's IMV minimum income scheme for a household of three; another reports that a pension rise now forces him to file a tax return.
-
Basque Country approves new rules to convert empty commercial units into housing
Decree 55/2026 lowers the minimum floor area to 35 m² and drops the dual-orientation requirement for converting empty commercial premises into homes, with price caps in stressed areas.
-
Spanish researcher evicted despite paying rent and holding a €71,000 purchase option
He paid rent for years, had a purchase option at €71,000 and was ultimately evicted by a Caixabank-owned fund, receiving a €35,000 compensation order.
-
The insurance was paid, but the goods weren't covered
A pipe bursts at 6:48 and soaks 32 boxes. The policy covered the building and a minimal amount of contents: the tenant's stock was not insured.