A lively discussion has unfolded around Anthropic's new model — Fable 5. The community is divided: some see it as a clever move to work with regulators, others as an expensive and inconvenient product, and still others as evidence of accelerating progress in the AI field. Let's break down three key perspectives shaping the picture around this model.
First Version: A Bet on Regulators
One expert, known in the crypto community as the creator of OutLayer, put forward the thesis that Fable 5 was created not for developers, but for governments and regulatory bodies. In his view, the model's strict limitations are not a bug, but a feature. Supporting this version is a package of accompanying announcements: early access for the state, joint reporting on "jailbreaks," a shared "severity scale" with Amazon, Microsoft, and Google, as well as federal AI safety teams.
This is not just a safety feature, but a bid to become a trusted partner of the state. According to the analyst's assessment, such investments are made not to sell tokens, but to "hold the pen when the rules are being written." Rules in the form of proprietary safeguards become a barrier that open models and small labs simply cannot overcome. The conclusion: the future of AI will be determined not by a benchmark, but by whom the regulator lets into the decision-making room.
Second Version: A Dubious Purchase for Developers
Another view, expressed by an analyst under the handle CyrilXBT, focuses on practical utility for developers. And here, the picture is bleak. Fable 5 costs $10 and $50 per million input and output tokens — exactly twice as much as its predecessor, Opus 4.8. Meanwhile, routine programming and debugging are now often blocked by a new safety filter, redirecting the user to a cheaper model.
Essentially, the user pays twice as much, waits for the request to be checked by a classifier, and receives a response from a cheaper model. Against this backdrop, CyrilXBT points to an alternative: for example, GLM-5.2 scores 74.4% on the FrontierSWE test, surpassing GPT-5.5 and trailing Opus 4.8 by less than one point, with an API cost six times lower. The model already works inside Claude Code by changing the base URL.
Third Version: A Symbol of Technological Breakthrough
A third perspective was offered by analyst Miles Deutscher, who focuses not on price or strategy, but on progress. According to his data, models of the Mythos class, to which Fable 5 belongs, have dramatically accelerated development. For six years, the length of tasks that AI can perform doubled approximately every seven months. With Mythos-class models, doubling occurs approximately every four months.
The dynamics are impressive: GPT-2 in 2019 handled tasks lasting about 3 seconds, GPT-4 in 2023 — about 4 minutes, Claude Opus 4.5 in December 2025 — about 5 hours, and Fable 5 in June 2026 — about 30 hours. Deutscher admits he cannot imagine where the industry will be in a year. This is a pure demonstration of exponential growth.
My analysis: Fable 5 is perhaps the most striking example of how three different realities collide in a single product: the ambitions of a lobbyist, the disappointment of a user, and the admiration of a techno-optimist. For the crypto industry, which itself balances between innovation and regulation, this case is an excellent illustration of how technology can simultaneously serve as a tool of influence, a commodity, and a milestone of progress. The question "who is Fable 5 created for?" actually leads us to a deeper one: what will the AI market look like when the rules of the game begin to be dictated not only by benchmarks, but also by regulators?