On June 9, Anthropic released the Claude Fable 5, the first Mythos-level model deemed safe for everyone. Prior to this, Mythos technology has been locked up in a closed program since spring because of too strong cyber capabilities. Now it is given to developers and subscribers, but with protective restrictions. In parallel, Claude Mythos 5 was released for a narrow range of partners. Let’s analyze in order: what is this model, which shows in practice how much it costs and what is the catch for business.
What is Claude Fable 5?
Claude Fable 5 is the public flagship of the new generation Claude 5. According to Anthropic, its capabilities surpass any model that the company released in the open access. It shows top results on almost all benchmarks tested: development, analytics, vision work, scientific research. The pattern is simple: the longer and more complex the task, the more noticeable the separation of the Fable 5 from other models.
The main thing to understand from the start is that Fable 5 and Mythos 5 are the same basic model. The only difference is the fuses. Public Fable has them on, partner Mythos has them on cybersecurity. Even the names refer to this: Fable from the Latin fabula, “what is told,” is related to the Greek mythos. It is the limitations that distinguish the models, so the names are different.
Where did the model come from: a brief chronology
To understand the scale of the release, you need context. The existence of Mythos became known in March 2026 due to the leak of drafts of the Anthropic blog. The company then confirmed the development and called it a qualitative leap. In April, Mythos Preview was launched privately through Project Glasswing, only for organizations dedicated to cybersecurity and critical infrastructure protection. The model was not allowed to open access: its ability to find vulnerabilities and collect exploit chains was considered too dangerous.
A few days before the official release, the identifier of the third-party aggregator lit up in the price list claude-oceanus-v1-p. It was an early preview checkpoint of the same model that was handed out to testers. The full chronology of that leak we collected in piecemeal.
What Fable 5 shows in practice
Anthropic brought specific cases of early customers, and they say more dry interest.
Stripe reported that Fable 5 compressed months of engineering work into days. In the 50 million-line Ruby codebase, the model migrated the entire database in one day, taking more than two months to manually migrate the team. At the Cognition FrontierCode benchmark, which tests quality supported code, Fable 5 performed best among frontier models, even at an average effort level.
In working with vision, the model sets a new bar. It restores the source code of the web application in one screenshot and passes Pokémon FireRed on minimal vision-harness – where the previous models Claude stalled even with auxiliary tools.
Memory and long context are noted separately. In Slay the Spire, file memory access improved Fable’s score three times as much as Opus 4.8, and it reached the final act three times as often.
How Fable differs from closed Mythos: fuses
The main difference between the public version is the classification system. These are individual AIs that catch potentially dangerous requests. When the Fable classifier is triggered on a topic in three categories—cybersecurity, biology and chemistry, or attempted distillation—the answer is automatically transmitted to the less powerful Claude Opus 4.8, and the user is notified.
According to Anthropic, such a rollback is rare: more than 95% of Fable sessions run without switching. For these sessions, its performance is virtually equal to a full-fledged Mythos 5. Fuses are deliberately configured strictly, so sometimes harmless requests fall under them. The company acknowledges this and promises to reduce false positives after launch.
Why is it so hard? Mythos-level models can find and exploit vulnerabilities, making cyberattacks cheaper and easier. In biology and chemistry, the logic is the same: the same queries are useful to the researcher and dangerous in the hands of an attacker. So for now, Fable is rolling back to Opus 4.8 on most queries from these areas.
Anthropic did an external bug bounty — more than 1,000 hours of testing — and in that time, no one has found a one-size-fits-all way around protection.
How much does a Claude Fable 5 cost?
The price of both models through the API is the same:
Entrance: $10 for 1 million tokens
Exit: $50 for 1 million tokens
That’s twice the price of the Opus 4.8, which has $5 and $25. But less than half the price of the former Mythos Preview, which cost $25 and $125. That is, Anthropic not only opened the Mythos class to the public, but also made it significantly cheaper than it cost in the closed version.
That's the order for subscribers. From June 9 to 22, Fable 5 is included in the Pro, Max, Team and seat-based Enterprise rates at no extra charge. From June 23, the model will be removed from these rates, and then use credits will be required. When capacity is enough, Anthropic plans to return the Fable 5 to standard rates. The API and consumption-based Enterprise model is available from day one.
Important for developers
If you’re working with the Claude API, there’s a technical detail that most retellings don’t have. The Fable 5 API surface is the same as Opus 4.7 and 4.8, but with one breaking change: the explicit parameter. thinking: {type: "disabled"} Now it returns error 400. If you don’t need an advanced reasoning mode, just don’t pass this parameter, rather than turning it off explicitly. Model identifier in the API - claude-fable-5.
Another point that concerns business directly. For all models of the Mythos level, mandatory storage of data for 30 days is introduced – on all sites, including third-party ones. This data is not used for training and is deleted after 30 days, but Anthropic needs it to protect against complex attacks and reduce false positives of classifiers. For those who have strict requirements for data processing, this should be included in the assessment.
What it changes in practice
In short, the Mythos level model is available for the first time to regular teams, not just special partners.
Multi-step agent scenarios get a more reliable basis. Opus 4.8 already holds long chains of tasks well, but Fable 5 is noticeably better at keeping context and plan at millions of tokens. Working with large codebases is accelerating, which is confirmed by the Stripe case.
At the same time, expect that in sensitive areas, the model will fail harder and switch to Opus 4.8. For legitimate tasks, this is rarely a problem, but if your work lies in the area of cybersecurity or biology, some of the requests will go away.
Frequently asked questions
What is Claude Fable 5?
It is the first public Anthropic Mythos model and the first in the Claude 5 lineup. Released on June 9, 2026, available through the Claude API and paid subscriptions.
How much does Claude Fable 5 cost?
10 dollars for 1 million incoming tokens and 50 for 1 million outgoing. This is twice the price of the Opus 4.8 and less than half the price of the closed Mythos Preview.
How is Fable 5 different from Mythos 5?
It's the same model. Fable has fuses for the public, Mythos has them removed in the field of cybersecurity – Mythos is available only to Project Glasswing partners.
What are the limitations of Fable 5?
In cyber security, biology, chemistry, and distillation attempts, the model does not respond by itself, but transmits the Opus 4.8 query. This works on average in less than 5% of sessions.
Which is better: Fable 5 or Opus 4.8?
Fable 5 shows top results on almost all benchmarks and in the case of Stripe for the day conducted a migration of the code base by 50 million lines. But it costs twice as much, so for simple tasks, Opus 4.8 remains a rational choice.
Outcome
Claude Fable 5 is the moment we’ve been waiting for since spring: Mythos’ technology has gone beyond closed programs, and it’s also cheaper than it cost in the closed version. With fuses in cybersecurity, biology and chemistry and priced at twice the price of Opus, but in the hands of conventional developers and subscribers. This is the first model of the Claude 5 line, and it raises the bar markedly. Next up is the OpenAI response with GPT-5.6.
