GPT-5.6 Sol got split and restrained, Fable 5 returned globally only after a 72-hour ban with its hands tied. Anthropic and OpenAI's strongest models both got 'brain surgery'.
Yesterday, OpenAI's top-tier series GPT-5.6 suddenly went live.
But GPT-5.6 was split into three forms:
- GPT-5.6 Sol: The next-generation frontier model,
- GPT-5.6 Terra: A balanced model suitable for efficient daily work,
- GPT-5.6 Luna: A fast and economical model for high-throughput tasks.
However, the release authority for the strongest GPT-5.6 Sol is no longer in OpenAI's Sam Altman's hands, but lies within an 'approved list'.
Before OpenAI conceded, Anthropic just experienced a 'ban crisis': Fable 5 was disabled globally except for some partner institutions.
Algorithm Rationing Era: GPT-5.6 Suffers 'Brain Surgery'
In the past, products were recalled when problems emerged. This time, the product was stopped at the door before even leaving the house.
According to the insider, OpenAI was asked to limit the release scope of GPT-5.6.
The reason is to restrain the strongest few models before the nationwide AI safety framework is implemented.
It was revealed that this intervention was due to GPT-5.6 possessing 'Mythos-like' capabilities, rather than a sudden adoption of more forceful intervention measures.
After careful consideration, OpenAI decided to restrict access to the latest AI, allowing only a limited number of users to preview GPT-5.6.
The vast majority of ChatGPT users will still need to wait several weeks for GPT-5.6 to be gradually unblocked.
Mythos 5 is Scary! Will Fable 5's Return Be a Weaker Version?
Before OpenAI conceded, another giant, Anthropic, had just experienced a crisis.
Two weeks ago, Anthropic released the mythical-level AI model Fable 5.
However, this model survived on the shelf for only 72 hours before being forcibly taken down.
Because of fear. Fable 5's cyber-offensive capabilities are frightening.
US House Representative Andrew Garbarino expressed it this way.
In a closed-door demonstration, Anthropic's Mythos demonstrated astonishing threats to the software security world, such as being able to empty private bank accounts.
Andrew Garbarino said:
Anthropic instructed the model to find vulnerabilities in the banking system and empty accounts, and then it went and executed it.
Subsequently, Mythos 'could also find this vulnerability and fix it'.
This was an unprecedented 'pull-the-plug' action in human history.
Only recently was Mythos 5 allowed to be opened to a very few 'trusted partners' while 'wearing shackles'.
According to the latest report: The 'castrated version' of Mythos 5, Fable 5, is expected to return and go live next week.
However, Fable 5 may be severely castrated to ensure the safety guardrails are more robust.
Right after Fable 5 was released, users found that simply saying the word 'cancer' would get them blocked by Claude.
And if the guardrails are further strengthened, the user experience might be a nightmare.
Shopify CTO Mikhail Parakhin has deeply tested Fable 5 and GPT-5.6 Max.
He believes Anthropic's absolute advantage is gone, GPT-5.6 has already caught up to Claude Mythos.
GPT 5.6 is clearly better than Opus 4.8 in all aspects (slightly faster too, though depending on the load).
Compared to Fable 5, it's clearly inferior in coding, but performs better on agent workloads.
This is a rare real-world comparison between Fable 5 and GPT-5.6, but it's enough to see that Anthropic cannot rest easy yet.
If Fable 5 returns only as a castrated version, it undoubtedly leaves more opportunities for competitors to catch up.
Are You Buying a Genius or a Puppet?
What makes developers most anxious is not just 'can't buy it', but 'getting a product that doesn't match the description'.
In AI circles, a term is spreading like a virus: 'Lobotomized'.
To pass those complex safety reviews, AI giants have no choice but to install layers of reinforced 'guardrails' on their models.
But intelligence often emerges from chaos and freedom.
When you forcibly castrate a model's ability to explore underlying code, encryption algorithms, or even deep logic, what you get may no longer be an omniscient 'Sol', but a mediocre 'Luna' with its claws clipped, obedient and submissive.
What you want is 'Prometheus's fire', but the US only allows you to export 'lighters'.
Developers generally fear that models will be heavily fortified with guardrails, even 'lobotomized', their capabilities castrated, severely reducing their intelligence.
Worse still, it might be limited to US use only, or require strict identity verification, instead of the originally promised 'wide availability, pay and use'.
Then there's the money issue.
Anthropic initially provided Fable 5 for free across multiple paid Claude subscriptions, open until June 22nd, giving users a brief window to experience its power—then, access vanished.
Now no one can say for sure: Will subscription users get back the promised 'free and unlimited use'? Or will it return, locked behind a door of extra charges or identity verification?
Maybe soon, we will miss that spring of 2023. Back then, as long as you had an email, you could commandeer a 'god-like AI'.
References:
https://punchbowl.news/article/tech/garbarino-mythos/
https://www.axios.com/2026/06/27/anthropic-fable-5-return-soon
https://www.wsj.com/tech/ai/openai-limits-access-to-new-model-citing-government-security-concerns-66420050
This article is from the WeChat public account 'New Zhiyuan', author: ASI Apocalypse
