Anthropic has launched Claude Fable 5.2 and Claude Mythos 5.2, its latest artificial intelligence models designed for complex coding, scientific research, cybersecurity and advanced technical work. The release brings faster responses, improved performance and lower API prices, while access to Mythos remains restricted to approved organisations.
Announced on November 9, the two models build on Anthropic’s previous Fable 5.1 and Mythos 5.1 releases. The company says the new models generate responses up to 30% faster and cost 20% less per token than their predecessors.
The launch also introduces a clear difference in access. Fable 5.2 is available to users through eligible Claude plans and developer platforms, while organisations working on approved cybersecurity and biomedical research projects can apply to use Mythos 5.2 through Anthropic’s verification programmes.
Anthropic says the models deliver improvements across software development, computer use, mathematics, robotics and scientific research.
What is new in Claude Fable 5.2 and Mythos 5.2?
Anthropic says both models share the same underlying model, but their safeguards and access arrangements differ.
Fable 5.2 is designed for demanding technical tasks and includes safeguards intended to reduce the risk of misuse in cybersecurity and biology. These protections allow Anthropic to make it available to a broader range of users.
Mythos 5.2 is intended for organisations that need more advanced capabilities for cybersecurity engineering and biomedical research. Access is limited to organisations approved through Anthropic’s Cyber Verification Program and Life Sciences Verification Program.
The company says the models can handle complex tasks with less human intervention than earlier versions. In its own testing, Anthropic reported progress in software migrations, hardware design, robotics and difficult mathematical problems.
One example involved a team of 1,000 Fable 5.2 agents working together to formalise a proof of Fermat’s Last Theorem in 30 hours. Anthropic said an earlier research version of Claude took 11 days to complete a similar effort with a purpose-built system and human guidance.
Formalisation means translating a mathematical proof into a format that a computer can check step by step. The result demonstrates how AI agents can divide a complex task into smaller jobs and coordinate their work, although the reported achievement reflects Anthropic’s specific experimental setup.
Anthropic cuts API prices by 20%
The release also makes the models cheaper for developers and businesses that access them through an application programming interface, or API.
The standard API prices are now:
- Input: $8 per million tokens, down from $10.
- Output: $40 per million tokens, down from $50.
Tokens are small units of text or other information that AI models process. Input tokens cover the information sent to a model, while output tokens cover the responses it generates.
Anthropic has also introduced an optional fast mode designed to generate tokens up to 2.5 times faster. This mode costs $16 per million input tokens and $80 per million output tokens.
The lower standard prices could make the models more affordable for developers building AI-powered applications, coding tools and automated business workflows. However, total costs will still depend on the volume of information processed, the length of responses and the level of reasoning required.
Anthropic says Fable 5.2 can also complete some demanding coding tasks at a lower cost than its previous models. In early testing on its FrontierCode 2.0 evaluation, the company reported that Fable 5.2 at high reasoning effort performed at a similar level to Opus 5.5 at maximum effort for roughly half the cost per task.
These are Anthropic’s own benchmark results and should not be treated as independent proof that the model outperforms every competing AI system.
What can the new models do?
Anthropic highlighted several areas where it says the new models have improved.
Software development: Fable 5.2 can work on complex codebases, identify bugs, improve performance and manage longer software projects. Anthropic said a group of agents using an earlier version of the model migrated a production service from Python to Rust, reducing the service’s startup time by about 26 times.
Cybersecurity: Mythos 5.2 is designed to identify software vulnerabilities and help security teams reproduce and fix them. Anthropic reported that its internal testing found more vulnerabilities with Mythos 5.2 than with Opus 5.5 and Mythos 5.1. Because these capabilities could also be misused, access is restricted through its verification programme.
Scientific research: The models have been used for difficult problems involving theoretical physics, protein structures and mathematical reasoning. Anthropic said Fable 5.2 solved two open problems involving black holes, while Mythos 5.2 helped develop a method for predicting the structures of complex viral protein shells.
Robotics and hardware: Anthropic reported improvements in simulated robot control and hardware design. In one experiment, Fable 5.2 helped design three working electronic devices, including a home voice-assistant hub and a USB speakerphone, with human engineers involved in building and testing them.
These examples illustrate the range of tasks Anthropic is targeting. They are not a guarantee that the models will produce equally successful results in every real-world setting.
Mythos remains restricted despite the launch
The limited availability of Mythos 5.2 is one of the most important parts of the announcement.
Anthropic says Mythos has stronger capabilities for finding and reproducing software vulnerabilities. Those skills can help cybersecurity professionals identify weaknesses before attackers exploit them, but they could also be used to support malicious activity.
Organisations pursuing approved cybersecurity or biomedical research projects can apply through Anthropic’s verification programmes. The company has not made Mythos 5.2 a general-purpose option for every user.
Fable 5.2, meanwhile, includes cybersecurity and biology safeguards designed to reduce the risk of misuse. Anthropic says these protections are as robust as those in Fable 5.1 while producing fewer unnecessary blocks.
The distinction reflects a broader challenge for AI developers: making more capable models useful to researchers and businesses without giving unrestricted access to capabilities that could create serious security risks.
How to access Claude Fable 5.2
Developers can access Fable 5.2 through Anthropic’s Claude Platform using the model identifier claude-fable-5-2. The model is also available through major cloud platforms, including Amazon Web Services, Google Cloud and Microsoft Azure.
Access through Claude’s consumer and business products depends on the user’s subscription plan and applicable usage limits.
Organisations seeking Mythos 5.2 must apply through the relevant cybersecurity or life sciences verification programme.
Anthropic recommends that users begin with Claude Opus 5.5 for most everyday coding and knowledge-work tasks, then move to Fable or Mythos when a project requires more advanced technical reasoning.
What the launch means for the AI market
The release puts pressure on AI developers to deliver more capable models without continually increasing the cost of using them.
For businesses, faster responses and lower API prices can make it more practical to use AI for complex coding, research and automated workflows. For developers, stronger reasoning and coding capabilities could reduce the time spent solving difficult technical problems.
However, performance claims need to be assessed carefully. Anthropic’s published benchmarks offer evidence of improvements on selected tests, but independent evaluations and real-world experience will help establish how consistently the models deliver those gains.
The launch also reinforces the growing importance of access controls as AI models become more capable in sensitive areas such as cybersecurity and biological research.
For now, Fable 5.2 gives eligible users a faster and cheaper option for demanding technical work, while Mythos 5.2 remains limited to approved organisations. How much the new models change everyday AI use will depend on their performance outside company-run tests, their cost at scale and the practical value they deliver to users.
Read also: Coinbase receives full US federal approval for trust bank