The second, Claude Mythos 5, is the same model but with unlocked cybersecurity capabilities – and that’s why it won’t reach the average user, but only a narrow group of trusted institutions cooperating with the United States government. This dual premiere says more about today’s artificial intelligence than might seem: the latest models have become so powerful that their full version is treated as dual-use technology.
One model, two names, one difference
The easiest way to understand the launch is this: Fable 5 and Mythos 5 are the same engine under the hood. They differ in only one thing – security. Fable 5 has a built-in protection system that, for questions in the areas of cybersecurity, biology, and chemistry, or attempts to copy the model, redirects the answer to the older, also very capable Claude Opus 4.8. Mythos 5 has these restrictions removed in the cyber sphere – which is why it remains closed within a trusted access program. The name itself reflects this relationship: “Fable” comes from the Latin word fabula, “that which is told,” related to the Greek word mythos. These are two sides of one story, differing only in their security measures.
Both models belong to a new class that Anthropic calls “Mythos” – a tier set above the previous Opus class. The first representative of this class was Mythos Preview, made available in April exclusively to selected cyber defense specialists as part of the government’s Project Glasswing program. Fable 5 is the first time the power of this class reaches the average user – albeit in a version with activated safeguards.
What it can do – specifics, not marketing
The manufacturer provides examples that are impressive even against the rapid pace of development in this industry. In tests conducted by the payment company Stripe, the model was supposed to compress months of engineers’ work into a few days: in fifty million lines of Ruby code, it performed a migration in one day, while it would have taken a human team over two months. In tasks requiring “vision,” the model recreated the code of a web application solely based on screenshots, and completed the game Pokémon FireRed using only raw screen images – without maps and hints that earlier versions needed.
Even more interesting things – if the data provided by the manufacturer is to be believed – are happening in science. In the Mythos 5 version, used by Anthropic’s internal experts, the model was said to accelerate certain stages of drug design by about tenfold, independently performing tasks that normally belong to a scientist. In genomics, after more than a week of largely independent work, it was supposed to build a machine learning model that – despite being a hundred times smaller – outperformed a solution previously published in the prestigious journal “Science.” Anthropic announces that it will publish the results of these studies in the coming months – until then, they remain a company declaration, although if confirmed, they will herald a moment when artificial intelligence ceases to be an auxiliary tool and becomes an active participant in research.
Cybersecurity – the heart of all caution
The most important aspect of this launch, especially for anyone thinking about online security, concerns cyber capabilities. According to Anthropic, Mythos 5 has the strongest cybersecurity capabilities of any model in the world. This is a double-edged sword. In the hands of defenders – those who secure critical software and infrastructure – such a tool can detect vulnerabilities and patch them faster than anyone before. In the hands of an attacker, these same capabilities would lower the cost and entry barrier for serious cyberattacks. This is why the full version of the model is not publicly available, and the version for everyone has aggressively set safeguards.
The company states that its safeguards have passed external resilience tests – according to Anthropic, over a thousand hours of penetration attempts within a bug bounty program did not yield a universal way to bypass protection, although the British AI security institute reportedly made some progress in this direction. This is an honest statement: not “we are unbreakable,” but “we strive to make breaking us slow and costly enough that we can detect it.” In the world of cybersecurity, such humility is usually a sign that someone understands what they are dealing with.
What this means for the average user
For the average user – including in Polish and Polonia companies, which increasingly use such tools – several things have practical significance. First, Fable 5 is available immediately, on all Claude platforms, and developers can use it via the programming interface under the name claude-fable-5. Second, in less than five percent of sessions, a user’s question will be redirected to the older Opus 4.8 model for security reasons – in the remaining more than ninety-five percent of cases, the conversation takes place at the full power of the new model. Third, the issue of costs and access: in the Pro, Max, Team, and Enterprise subscription plans, the model is available without additional charges only until June 22; from June 23, its continued use in these plans will require additional credits until the company increases computing power enough to restore it as a permanent part of the subscription.
The price for developers is ten dollars per million input tokens and fifty per million output tokens – less than half of what the previous Mythos Preview cost. In practice, this means that the most powerful publicly available artificial intelligence to date is also becoming cheaper. This is a trend that has been repeating in this industry for years: capabilities grow, and prices fall.
A broader lesson
The most interesting thing about this launch is not that the new model is better than the previous one – that’s the norm in the AI world. What’s interesting is that Anthropic has for the first time so clearly separated one model into two versions solely for security reasons, treating the full extent of its capabilities as a technology that cannot simply be given to everyone. This is a signal that the industry is entering a phase where the question “what can the model do” is inseparable from the question “who is allowed to use it.” For the user, the good news is that they get a tool more powerful and cheaper than ever. The question that remains open – and which the industry will have to ask itself more and more often – is: how long can the line be maintained between what is public and what is too powerful to be made public.
Key facts
- What: Anthropic released Claude Fable 5 (a publicly available model) and Claude Mythos 5 (the same model without some cyber security features, only for trusted government partners). Launched June 9, 2026.
- Key: Fable 5 is Anthropic’s most capable publicly available model; for questions in cybersecurity, biology, and chemistry, it redirects to Opus 4.8 (below 5% of sessions). Price: 10/50 USD per million tokens.
- Source: Anthropic press release dated June 9, 2026.
Arthur Skok, poland.us editorial team. Based on Anthropic press releases. More about technology and artificial intelligence from the perspective of the Polish diaspora at poland.us. Polish diaspora directory of tech companies and services: PolishPages.com.
📱 Download the PolishPages app:
App Store (iPhone) | Google Play (Android)
