Anthropic officially releases Claude Fable 5 and Claude Mythos 5. The former is open to ordinary users, while the latter continues to be restricted to "trusted security partners." The naming of the new model can be said to be in line with Claude’s consistent style.
From Haiku (haiku) to Sonnet (sonnet) to Opus (artistic masterpiece), Anthropic has been layering models with literary and artistic concepts. By Mythos, the name has expanded from literary works to "mythology" itself.
Fable comes from the Latin fabula, which means "something that is told", and has the same origin as the Greek mythos. When translated literally, it is generally called a "fable". The name is just like the positioning of the new model, a "Mythos-level" model, an "open myth".
According to Anthropic's description, Fable 5 and Mythos 5 share the same underlying model, but are packed into a secure shell more suitable for public distribution. In terms of official model capabilities, the two of them are placed in the same position.
But scores are scores, and if Fable and Mythos performed exactly the same, I think there would be no need to separate them into two names.
The rewritten "myth"
Myths were rewritten, compressed, and admonished, and became fables.
According to the official documentation, Fable 5 is a public version. It is open to ordinary users and developers, but in high-risk areas such as network security, biology, chemistry, and model distillation, additional security classifiers will be involved. Once the system determines that the request may involve these sensitive directions, the answer will not be continued by Fable 5, but will automatically fall back to Claude Opus 4.8.
Mythos 5 is based on the same underlying model, but lifts Fable 5's guardrails in some areas. Anthropic said that network security partners in Project Glasswing can use the "full version" of Mythos 5; in the future, some life science researchers may also use versions that remove biological and chemical restrictions through the Trusted Access Program.
We won’t mention the unused Mythos for now, let’s look at some practical things first.
The first is pricing. In one word, expensive.
The pricing of Fable 5 is $10/million tokens for input and $50/million tokens for output. Developers can now call claude-fable-5 (model name) through the Claude API.
This price is exactly twice that of Opus 4.8, and it is the same as the fast mode of Opus 4.8. Anthropic has obviously placed it in a higher price tier than Opus.
However, Anthropic said that this price is less than half of the previous Claude Mythos Preview - but since Mythos Preview is not a public API model, the official has not given a standard price for the public, and this sentence cannot be verified.
Subscribers also need to note that Fable 5 may not be directly included in the basic subscription package for a long time.
Anthropic mentioned in its official description that after June 23, even if users have subscribed to Claude, Fable 5 may be provided on a volume basis based on computing power and may not be directly included in the basic subscription service.
This company is becoming more and more stingy, but at least they still have a trial period of half a month. The official also left some leeway: if computing power resources are sufficient after June 23, Anthropic will try to continue to include Fable 5 in Pro, Max and other subscription services.
The high price is not difficult to understand in itself, but it is best to ensure that its capabilities are worthy of its price.
Judging from the running scores, Fable 5/Mythos 5 is basically Anthropic’s strongest one currently.
However, there is a note in the official table that the scores of Claude Fable 5 and Claude Mythos 5 are generally only 1-3 percentage points apart (except for network security and bio-related tests with asterisks), so the table shows the higher score of the two. It's hard not to complain about this.
Anthropic focuses Fable 5 on several directions: software engineering, knowledge work, vision, long context memory, and life science research.
Software engineering is one of the most prominent scenarios. According to the table, Fable/Mythos 5 reached 80.3% on SWE-Bench Pro, which is significantly higher than Opus 4.8's 69.2%; on the more difficult FrontierCode Diamond, it got 29.3%, while Opus 4.8 only had 13.4%, and GPT-5.5 had only 5.7%.
For knowledge work and visual tasks, Anthropic gives two types of evidence.
One type is standardized benchmark. The official table shows that Fable/Mythos 5 scored 1932 on GDPval-AA, higher than Opus 4.8, GPT-5.5 and Gemini 3.1 Pro; on the GDP.pdf document task with visual understanding, it reached 29.8%, also surpassing other major models.
Another category is early customer testing. Anthropic said that Fable 5 achieved the highest score on Hebbia's advanced financial reasoning benchmark, with its advantages focusing on document reasoning, chart and table understanding, and problem solving; IMC also reported that it almost fully passed the transaction analysis assessment.
In order to demonstrate the visual capabilities of Fable 5, Anthropic gave an example: Previously, the Claude model required complex auxiliary tools to play "Pokémon: Fire Red", but Fable 5 can complete the level with only visual input.
In terms of long-term tasks and memory capabilities, Anthropic said that Fable 5 can stay focused on long-term tasks with millions of tokens and use its own notes to improve output.
In games like "Slay the Spire" that require continuous decision-making and long-term strategy, if Fable 5 is connected to persistent file memory so that it can record previous choices and experiences, its performance will be significantly improved. The improvement is three times that of Opus 4.8, and the number of times you can reach the final level is also increased three times.
By the way, Fable is also the name of a classic RPG game, translated into Chinese as "Fable" - maybe one day we can see Fable playing "Fable".
In addition, in terms of network security capabilities, Fable/Mythos 5 reached 78.0% in ExploitBench Cap%, exceeding Claude Mythos Preview's 69.0% and nearly twice that of Opus 4.8.
The score of Mythos 5 should be used here, because Fable 5 will fall back to Opus 4.8 on high-risk requests.
Strong models must be layered
This time, Anthropic put the visual display of the model's capabilities into a sliding page similar to a "portfolio", and only gave a short paragraph of comments for each demo.
For example, Fable 5 wrote a solar system simulation that deduced planetary orbital motion from first principles of physics and used it to predict solar eclipses.
For another example, it can play "Factory" independently. This is a factory automation game that engineers love. Players have to collect resources, plan production lines, and build logistics and energy systems.
Anthropic used this example to illustrate that Fable 5 can formulate strategies in an open environment and continue to promote the construction of a complex system.
In another demo, Fable 5 first made a browser-based CAD editor, and then used this self-developed CAD tool to design a complete model that can be 3D printed. This editor also has built-in AI copilot to assist modeling.
The focus of this demo is that Fable 5 completes a closed loop: first create the tool, then use the tool, and finally complete a physical design task.
The last demo, Anthropic showed off a fluid simulation written by Fable 5, with motion rhythm synchronized with a classical music EDM remix. The official also specifically mentioned that the music is also generated by Fable 5 using code.
These examples look fancy, but the idea is the same: Fable 5 is great at mixing code, vision, physics, design, and long-term planning into tasks.
If this part shows what Fable 5 can do in the hands of developers, the following part talks about what Mythos 5 can do in the hands of researchers - and why Anthropic wants to separate Fable and Mythos.
Anthropic said that in evaluations by internal protein design experts, Mythos 5 accelerated some aspects of the drug design process by approximately 10 times. In one case, Mythos 5 plugged into protein design and bioinformatics tools, without human help, to match or even surpass skilled human operators.
In this task, Mythos 5 does not simply answer questions, but completes a complete set of scientific workflows: selecting binding sites, selecting and running protein design tools, and recovering itself after failure. Officials said that of the 14 protein targets in this study, 9 produced strong candidate molecules and are currently undergoing further research.
Anthropic also mentioned that Mythos 5 can steadily propose novel and attractive molecular biology hypotheses. In a blind test comparison with Opus-level models, in-house scientists preferred the hypotheses proposed by Mythos about 80% of the time, some of which have been evaluated experimentally.
Meanwhile, one of Mythos 5's hypotheses about a new mechanism for the E. coli protein was confirmed in a study by another laboratory independently studying the same problem.
It even did a genomics study.
Anthropic said Mythos 5 completed a new genomics study almost autonomously in just over a week. It organized single-cell data from millions of cells across 138 animal species and designed and trained a custom machine learning model to identify cells that perform the same role in different species.
What’s even more exaggerated is that Anthropic claims that the performance of the model trained by Mythos 5 exceeds that of a model recently published in Science, although its scale is only one percent of the latter. Anthropic said it plans to publish these results in the coming months.
Of course, this part still needs to wait for the paper and external review. But if you only look at the official information provided by Anthropic, Mythos 5's capabilities in life sciences are close to that of a scientific research agent: it can read questions, use tools, process data, train models, propose hypotheses, and advance a research to the point where it can be published.
Once a model can truly advance tasks in drug design, viral vectors, protein design, and genomics research, it will naturally have dual uses.
It can be considered that the life science part is not an ordinary functional demonstration of Fable 5, but a demonstration of the upper limit of Mythos 5's capabilities.
But it is shown to illustrate that the underlying model in Anthropic's hands is so powerful that it must be distributed through trusted access.
Interestingly, model release in the past was more like a technical product: parameters, running scores, price, context length, API name. Strong is strong, release is release.
But here in Mythos, things start to get complicated. The same underlying model is divided into two versions. Ordinary users get Fable 5, while security researchers and some life science researchers get Mythos 5; the former is installed in stricter guardrails, and the latter requires trusted access.
Model capabilities are no longer only differentiated by high and low, but also by authority, scenarios, and responsibilities.
In a sense, it can be seen as a signal that the commercialization of cutting-edge models has entered a new stage - the stronger the model, the less likely it is to be thrown directly to everyone. They will be split into different versions, put into different security boundaries, and then given to different types of users.
Anthropic did this first because their model was “too strong to be fully disclosed.” In the future, other companies that want to tell stories and prove that their models are strong may also follow this approach.