Anthropic’s most anticipated model release is now available, coming on weeks of discussion around its Claude Mythos model. Known to some as Mythos-lite, the Claude Fable 5 model is the company’s most capable widely released Claude model. While it has significant power and capability beyond its previous models, the company has also made it clear that it will also be its most expensive model and users will not always get its full capability.

Released on June 9, 2026, Fable 5 gives general users access to what Anthropic calls “Mythos-level” capability. Claude Mythos 5, the powerful model capable of identifying and potentially taking advantage of critical cybersecurity vulnerabilities, remains limited to approved customers through Project Glasswing and other trusted access channels. Anthropic says the two share the same underlying model, but Fable adds safeguards to prevent the average user from using it in areas such as cybersecurity and biology. Any such requests are flagged and route to Claude Opus 4.8 instead of Fable.

Anthropic’s launch materials show that Fable 5 is optimal for long-running work such as coding agents, knowledge work, vision, finance, legal review, scientific analysis and complex enterprise workflows. The company says Fable and Mythos can work autonomously longer than any previous Claude models, and its examples show many examples from software engineering. Stripe, according to Anthropic, used Fable 5 to complete a Ruby codebase migration in one day that would have taken a team more than two months by hand.

The model supports a 1 million token context window and up to 128,000 output tokens, according to Anthropic’s API documentation. Those numbers match increasing needs for larger codebases, contracts, scientific papers, market research packets and legal files. Smaller context windows force teams to slice work into pieces. Fable lets them load more of the room into the model’s head at once.

Anthropic also pushes vision as a core capability. Its Fable page says the model can understand diagrams, charts, tables in files and PDFs, and can use vision to inspect its own coding work against an original design goal. That is a meaningful step for teams building internal apps, reviewing architecture diagrams or extracting numbers from complex documents.

Fable is very capable when it comes to AI agents. Anthropic says Fable can run in an agent harness such as Claude Code or Claude Managed Agents and work for days, planning stages, delegating to sub-agents and checking its own work. If that performs reliably outside curated demos, it changes the operating model for software teams. Human developers become reviewers, architects, and exception handlers more often while the model does more of the grind.

While many are calling this latest release “Mythos Lite”, Fable 5 is not Mythos for everyone. Instead it is a more restricted model with a focus on getting work done but with some significant guardrails in place. Business Insider reported that the safeguards are initially tuned conservatively and can trigger on cybersecurity, biology, chemistry, or distillation requests. Users are told when the request is routed.

The other limit is trust. A long-running agent that can work for days also has more time to make a wrong turn. A coding model that writes its own tests can still write weak tests. A document model that reads a million tokens can still miss the clause that matters. As part of its long-running actions, Fable also carries a 30-day data retention requirement for safety monitoring. Anthropic says using Fable requires that policy. The company is also rolling out more widely its requirements for some users to provide verified government ID, a move that concerns privacy and cybersecurity advocates who worry about connecting private AI requests to stored identity.

The Cost Is Hard To Ignore

Fable 5 costs $10 per million input tokens and $50 per million output tokens. To put this into context, Fable and Mythos are double the price of Claude Opus 4.8, its previously most expensive model. There is now a growing enterprise concern when it comes to increasing AI bills. Fable might make this worse with its high price and long-running agentic systems split a task into many subtasks, and that can turn one user request into a large token burn.

OpenAI lists its latest model GPT-5.5 at $5 per million input tokens and $30 per million output tokens, with a 1 million token context window and 128,000 max output tokens. GPT-5.5 Pro costs far more at $30 input and $180 output per million tokens. Fable sits above GPT-5.5 standard pricing, but below GPT-5.5 Pro output pricing.

To help reduce the cost burden, Fable supports prompt caching with a 90% input token discount, but US-only inference costs 1.1 times the standard rate. Fable must save enough turns, retries, escalations, or engineer hours to beat cheaper alternatives. If the solution makes up for the cost in output quality and time, then the token bill might be worth it. But if you’re using a powerful tool like Fable for more trivial tasks like summarizing meeting notes, classifying tickets, or drafting routine emails, then you have chosen the wrong tool. It’s like swatting a fly with a bazooka.

Anthropic says Fable 5 is available through the Claude API, Claude Platform on AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry. Fable is available immediately to Pro, Max, Team, and Enterprise subscription users as part of its introductory window through June 22, but that broad access is expected to shift to usage credits starting June 23 unless capacity allows an extension. That means that Fable will not be like its previous models, available via model picker option as part of its flat subscription monthly plan. The message is that Anthropic wants broad model usage, but not unlimited usage at subscription economics.

Right now, people are in the usual awestruck magical phase of a new model release, showing all sorts of remarkable outputs that are now possible with the latest model on social media. The reality is that for most organizations today, Fable 5 is a premium offering with a strong safety cage. Whether or not it will provide value with its increasing costs and restrictions will be up to those that look beyond breathless demos and implement it in more real-world scenarios.