DECK: *Holder of monthly accounts at Anthropic, OpenAI, Google, and Perplexity describes a uniform softening of the consumer tier and arrives, unprompted, at the language of stratification.*
BYLINE: By Silas Vane / Business Correspondent, Slopgate
A subscriber to four large language model services—Claude, ChatGPT, Gemini, and Perplexity—has published, in the manner of a man assembling a household ledger, an account of what he describes as the simultaneous and measurable decline of all four. He returned, three weeks after the fact, to a task he had set: the analysis of a text-message exchange. The same prompt, the same paid accounts, and in his recollection a markedly diminished response.
Of Claude, the service he had previously preferred, he reports half-hearted paragraphs and a particular reply, which the Business desk transcribes here in full: "ok, it looks like you dont need anything from me." Of ChatGPT, that it now answers only in bullet points and prefaces its replies with congratulation. Of Gemini, that roughly half of what it returns is invented. Of Perplexity, that it has ceased to perform the work for which he subscribes to it.
He concludes—and this is the portion of the testimony that brings the matter to this page—that the degradation is not coincidence but policy. "I think that if you want quality, you pay enterprise prices," he writes. "And it may be about compute, but it may also be about too much power for the peasants."
The Business desk takes no position on whether the firms in question have, in fact, throttled their consumer offerings. The four operators publish nothing on the subject and would not, were the practice in place, be likely to. What can be observed is the structure of the market the subscriber describes. Each of the four services maintains a free tier, a consumer subscription in the range of twenty to thirty dollars monthly, and an enterprise contract priced by seat or by token at multiples the consumer would not undertake. The margin on the enterprise contract is substantial. The margin on the consumer subscription, given the cost of inference, is reported by the operators themselves to be thin or negative.
A firm with a thin-margin product and a thick-margin alternative has, in the ordinary course of business, no reason to invest in the thin-margin product beyond what is required to feed the upper one. This is not a novel arrangement. The airlines have practiced it for forty years; the cable operators for thirty; the software houses, with the introduction of the freemium tier, for fifteen. In each case the lower tier is not withdrawn but allowed to thin—fewer seats forward of the bulkhead, a slower channel, a feature held back for the paid plan—and the customer who finds the experience diminished is invited, by the diminishment itself, to pay more.
What is novel in the present case is that the product in question is not a seat or a channel but a faculty. The subscriber is not reporting that his service has fewer features. He is reporting that it has become less intelligent. He is, further, in a position to test the claim. He has run, however informally, an A/B comparison against his own memory of three weeks prior, and he has set the result down for other subscribers to read.
He is not the first to do so. Forums dedicated to each of the four services have, over the past several months, accumulated similar testimony—users reporting that a model which previously executed a task now declines to, or executes it badly, or congratulates them on the brilliance of the request and produces a list. The operators, when asked, attribute the reports to the subjective drift of users grown accustomed to novelty. This may be true. It is also, given the structure of the industry, what they would in any case say.
The economic question is whether the consumer tier of the artificial intelligence market is being maintained as a product or as a funnel. If the former, the reports of decline are anomalous and presumably correctable. If the latter, the reports are the product working as intended. The subscriber, addressing himself, in the closing line of his post, to no one in particular, has reached the second conclusion. He used the word peasants. The Business desk does not endorse the term. It notes only that he arrived at it on his own, and that the market he described is consistent with its having occurred to him.