Claude Fable 5 Explained: Capabilities, Safety & What's Next | Agent5
AI Models & Releases
Claude Fable 5: Anthropic's Most Powerful Public AI, Explained
By The Agent5 founder·June 28, 2026
Claude Fable 5 is Anthropic's most capable publicly released model, bringing Mythos-class reasoning and agentic power to general users for the first time. Its launch and rapid suspension by U.S. export controls mark a turning point in how governments and AI labs negotiate the edge of frontier AI.
Key takeaways
Claude Fable 5 is Anthropic's most capable publicly released model, bringing Mythos-class reasoning and long-horizon agentic ability to general users for the first time, built on the same underlying weights as the restricted Claude Mythos 5.
Safety classifiers route high-risk requests in cybersecurity, biology, and chemistry to Claude Opus 4.8 instead, triggering in fewer than 5% of sessions and keeping the most dangerous capabilities out of general reach.
Benchmark performance shows substantial leads over prior Claude models and competing frontier models on software engineering, complex reasoning, vision, and long-horizon agentic tasks, with a 1 million token context window and pricing at $10 per million input tokens.
Three days after launch, the U.S. Commerce Department issued an export control directive forcing Anthropic to suspend both Fable 5 and Mythos 5 globally, marking what appears to be the first government-mandated takedown of a publicly deployed frontier AI model.
The Fable 5 episode is a concrete signal for thinking about the future: governments are developing export control frameworks for AI models, mandatory data retention is becoming a condition of accessing frontier capabilities, and fallback design is now a practical requirement for anyone building on top of cutting-edge AI.
Something rare happened on June 9, 2026. Anthropic released what it described as the most capable AI model ever made broadly available, and then, just three days later, a government directive forced it offline for all customers worldwide. The story of Claude Fable 5 is not just a product launch story. It is a live case study in how powerful AI gets built, guarded, released, and contested all at once. If you want to reason clearly about where AI is heading, understanding what Fable 5 is, what it can do, and why it was suspended gives you some of the sharpest signal available right now.
What Is Claude Fable 5, and Where Does It Fit?
Claude Fable 5 is a Mythos-class model that Anthropic has made safe for general use. That single sentence carries a lot of weight, so it helps to unpack the terminology.
The Claude family covers three named size tiers, Opus, Sonnet, and Haiku, introduced together with Claude 3 in March 2024, joined in June 2026 by the Mythos-class tier that sits above the Opus class, with the current frontier being Claude Fable 5, announced on June 9, 2026.
While Opus was optimized for general high-capability tasks, Mythos-class models are specifically designed with agentic use in mind: longer reasoning chains, better tool use, stronger instruction adherence across complex multi-turn conversations, and improved reliability when operating with limited human oversight.
The "Fable" name distinguishes the publicly available version from the restricted version. Anthropic claims that Claude Fable 5 offers the same performance as Claude Mythos 5, except with much more strict guardrails in place to prevent it from being used for harmful things. In other words, both products run the same underlying model weights. The difference lies entirely in what each version is allowed to do.
A Brief Timeline: From Mythos Preview to Fable 5
To understand Fable 5, you need to know its predecessor. Anthropic unveiled Mythos in April 2026 and limited the rollout because of its advanced cybersecurity capabilities. That earlier release, called Mythos Preview, was offered only to a small group of vetted organizations.
Launched as a preview in April, Mythos was initially limited to a handful of partners due to cybersecurity concerns. Last week before the Fable 5 launch, Anthropic expanded access to hundreds of organizations across 15 countries, again focusing on organizations that manage critical infrastructure.
With the launch of Claude Fable 5, Anthropic honored its stated "eventual goal" to deploy Mythos-class models at scale. The strategy was simple in concept: build guardrails strong enough that the most dangerous capabilities stay locked, then release the rest of the model's power to anyone who needs it.
What Claude Fable 5 Can Actually Do
Fable 5's capabilities exceed those of any model Anthropic has ever made generally available. It is state-of-the-art on nearly all tested benchmarks of AI capability, showing exceptional performance in software engineering, knowledge work, vision, scientific research, and many other areas.
The software engineering results are particularly striking. During early testing, Stripe reported that Fable 5 compressed months of engineering into days. In a 50-million-line Ruby codebase, the model performed a codebase-wide migration in a day that would otherwise have taken a whole team over two months by hand.
On independent coding benchmarks, the numbers back that up. On SWE-Bench Pro, Claude Mythos 5 and Fable 5 scored 80.3%, ahead of Claude Mythos Preview at 77.8%, Claude Opus 4.8 at 69.2%, GPT-5.5 at 58.6%, and Gemini 3.1 Pro at 54.2%. On FrontierCode Diamond, they scored 29.3%, compared with 13.4% for Claude Opus 4.8 and 5.7% for GPT-5.5.
Knowledge work shows similar gains. On Hebbia's Finance Benchmark for senior-level reasoning, Fable 5 has the highest score of any model, with substantial gains in document-based reasoning, chart and table interpretation, and problem solving. IMC noted that Fable 5 aced their trading-analysis evaluations nearly across the board, including factual lookup, conceptual reasoning, root-cause analysis, and expected-value analysis.
For analytical work, analytics company Hex said in a statement that Fable was the first to get a 90% on its core analytics benchmark of complex, long-running analytical tasks.
The longer and more complex the task, the larger Fable 5's lead over other models. That pattern matters for understanding where the model genuinely earns its cost premium.
The Agentic Leap: Working Autonomously for Days
One of the defining features separating Fable 5 from earlier Claude generations is its ability to run extended, unsupervised tasks.
Fable 5 and Mythos 5 can work autonomously for longer than any previous Claude models. These skills apply to software engineering and cover improved capabilities in knowledge work, vision, memory, and life sciences research.
The model can tackle days-long, complex, and asynchronous tasks that previous models couldn't sustain. When run inside an agent harness like Claude Code, it can work for days at a time, planning across stages, delegating to sub-agents, and checking its own output.
Vision capabilities also took a meaningful step forward. Claude Fable 5 understands diagrams, charts, and tables nested in files and PDFs. This opens up research and document-heavy work in finance, legal, analytics, architecture, and gaming. In coding, the model implements designs with high fidelity and uses vision to critique its output against goals.
For the life sciences, the results described in early testing are notable. The launch post and system card describe very strong Mythos 5 life-sciences results: drug-design acceleration, novel molecular-biology hypotheses, genomics research, and benchmark gains on BioMysteryBench, LatchBio Bioinformatics, structural biology, ProteinGym Hard, organic chemistry, protocol troubleshooting, and LABBench2.
The Guardrail Architecture: How Safety Is Built In
Releasing a Mythos-class model to the general public required Anthropic to solve an engineering problem it had never faced before: how do you prevent a model's most dangerous capabilities from being accessible while keeping everything else fully functional?
Without safeguards, Fable 5's capabilities in areas like cybersecurity could be misused to cause serious damage. Anthropic therefore launched the model with safeguards that mean queries on some topics will instead receive a response from its next-most-capable model, Claude Opus 4.8.
Claude Fable 5 includes safeguards that limit its performance in specific areas where misuse risk is elevated. Harmful prompts related to cybersecurity, biology, chemistry, and health fall back to receive a response from Opus 4.8 instead. Importantly, users are told when a fallback happens, and they are not charged Fable prices for rerouted requests.
The fallback triggers rarely for most users. Anthropic says the cases in which Fable has to defer to Opus 4.8 are rare, with early data showing at least 95% of Fable sessions running entirely on the model's own responses.
To validate the classifiers before launch, Anthropic worked with the US government, the UK AISI, multiple private third-party organizations, and internal teams to red-team Fable's safeguards for thousands of hours in total. These tests showed that Fable's safeguards are substantially more effective than those of any previously deployed model. No testers were able to find a universal jailbreak.
There is also a data retention requirement attached to using the model. With the launch of Fable 5 and Mythos 5, Anthropic requires a 30-day retention on all traffic, even if enterprises previously had zero-retention agreements. The company says it will not use the data for training and will use it only to "defend against complex and novel attacks, including new jailbreaks," and "identify and reduce false positives."
Pricing and Access
Fable 5 and Mythos 5 are being offered at $10 per million input tokens and $50 per million output tokens, less than half the price of Claude Mythos Preview. For comparison, that pricing is double the price of Opus 4.8.
Claude Fable 5 and Claude Mythos 5 share a 1M token context window by default, and support up to 128k output tokens per request. That context window makes it possible to load enormous codebases, long documents, or extended research threads into a single session.
Claude Fable 5 is generally available on the Claude API, Claude Platform on AWS, Amazon Bedrock, Google Cloud, and Microsoft Foundry. Claude Mythos 5, by contrast, is not generally available and is offered in limited availability to approved customers in Project Glasswing.
For subscription users, the initial rollout had a staged plan. Through June 22, Fable 5 was included in Pro, Max, Team, and seat-based Enterprise plans at no extra cost. On June 23, Anthropic planned to pull Fable 5 from those plans, requiring usage credits going forward, with plans to restore it as a standard subscription feature as soon as possible. Those plans were overtaken by events.
The Suspension: A Government Kill-Switch in Action
Fable 5 was available to the public for exactly three days.
Anthropic was forced to disable all access to its newest AI models, Fable 5 and Mythos 5, late on Friday after the U.S. Commerce Department used national security export controls to bar the company from distributing the models to any foreign national.
The directive includes not just people located outside the U.S., but also any foreign national in the U.S., including Anthropic's own non-citizen employees. Given the scope of the directive, Anthropic argued it had no choice but to disable the models for all users.
The stated reason for the suspension centered on a potential jailbreak. Anthropic's understanding is that the government believes it became aware of a method of bypassing, or "jailbreaking," Fable 5. Anthropic reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities.
Anthropic pushed back publicly. The company disagrees that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people, and argues that if this standard were applied across the industry, it would essentially halt all new model deployments for all frontier model providers.
This appears to be the first government-forced takedown of a publicly deployed frontier model. As of this writing, the model remains offline for all users globally, with Anthropic working to restore access.
The Agent5 Angle: Thinking in Probabilities About What Comes Next
The Fable 5 episode is dense with signal for anyone trying to reason about the near future of AI. Here are the questions worth holding in mind, and how to think probabilistically about the answers.
Will models like Fable 5 return to general availability? Almost certainly yes, and probably soon. Anthropic's public statement frames the suspension as a misunderstanding, not a fundamental policy disagreement. The export control regime is new, contested legal territory. Anthropic suspects that perfect jailbreak resistance is not currently possible for any model provider, and that every safeguard used in the industry is vulnerable to non-universal jailbreaks in specific circumstances. If that argument prevails in negotiations, the bar for suspension would need to rise substantially before it could be applied again. Assign a high probability to Fable 5 returning in some form, with additional verification requirements attached.
Will capability keep compounding? The launch follows Anthropic's plea urging major global AI labs to establish a coordinated brake pedal on frontier AI development. Anthropic warned that systems are advancing so rapidly that they may soon achieve recursive self-improvement, autonomously improving themselves without human intervention. That framing, combined with the gap already visible between Fable 5 and Opus 4.8 on agentic tasks, makes it reasonable to predict the next generation of publicly available models will be more capable still. The direction is clear even when the timing is uncertain.
Does the government's action reduce actual risk? This is the hardest question and the most instructive. Anthropic said it believed the same jailbreak could be used to elicit similar capabilities from other publicly available models, including OpenAI's GPT-5.5, that are not subject to similar national security export controls. Meanwhile, the US government suspended Fable 5 and Mythos 5 to prevent foreign nationals from accessing world-class autonomous security capabilities. Within 16 days, a Chinese lab operating outside US export control jurisdiction publicly claimed parity on those same capabilities. That sequence is worth factoring into any prediction about whether export controls on AI models are an effective long-term strategy.
What does mandatory data retention mean going forward? The 30-day retention policy could set an industry precedent in which access to increasingly powerful models comes with mandatory data-retention policies framed as a safety measure. Getting smart about AI means recognizing that the terms of access to frontier models are themselves a moving target, and that commercial, legal, and regulatory conditions will increasingly shape what developers can build and how.
For anyone building on top of AI today, Fable 5 illustrates a new kind of infrastructure risk: the model you depend on can be switched off by a government directive faster than your incident response team can convene. That is not a reason to avoid powerful models. It is a reason to design systems with fallback paths in mind from the start.
Yes, they share the same underlying model weights. The difference is in access and safeguards. Fable 5 is the publicly available version and includes safety classifiers that route certain high-risk requests in cybersecurity, biology, and chemistry to Claude Opus 4.8. Mythos 5 has those classifiers lifted and is restricted to a small group of vetted customers through Project Glasswing.
On June 12, 2026, just three days after launch, the U.S. Commerce Department issued an export control directive ordering Anthropic to suspend access to Fable 5 and Mythos 5 for any foreign national, whether inside or outside the United States. Because Anthropic could not filter users by nationality in real time, it disabled the models for all customers globally. The government cited a reported jailbreak of Fable 5's safeguards; Anthropic disputed that the jailbreak was broad enough to justify a full recall.
If a request to Fable 5 touches topics like offensive cybersecurity, dual-use biology, or chemistry, the model's classifier automatically routes that request to Claude Opus 4.8 instead. The user is notified that this happened and is not charged Fable 5 pricing for the rerouted response. Anthropic's early data suggests this fallback is triggered in fewer than 5% of sessions on average, so most users never encounter it.
Substantially better on long, complex tasks. On SWE-Bench Pro, Fable 5 and Mythos 5 scored 80.3%, well ahead of Claude Opus 4.8's 69.2% and competing frontier models. Stripe, one of the early testers, reported the model completed a codebase-wide migration of a 50-million-line Ruby codebase in a single day, a task that would have taken a whole team more than two months by hand.
Claude Fable 5 comes with a 1 million token context window by default and supports up to 128,000 output tokens per request. Pricing is $10 per million input tokens and $50 per million output tokens. A 90% discount applies to prompt-cached input tokens. US-only inference is available at a 1.1x price premium.
It may. This appears to be the first government-forced takedown of a broadly deployed frontier AI model. The export control directive was broad enough to require Anthropic to disable the model for all users worldwide, not just foreign nationals. Anthropic publicly argued that applying this standard across the industry would essentially halt all new frontier model deployments, suggesting the company intends to contest the precedent rather than accept it quietly.
