Anthropic Releases First Mythos-Class AI Model To The Public

Anthropic has released its first Mythos-class artificial intelligence model to the public, marking a significant step in the company’s cautious rollout of more powerful AI systems with advanced reasoning and software capabilities.

The company announced Claude Fable 5 on Tuesday, positioning it as its most capable widely released model to date. The new model is designed for demanding reasoning, long-horizon agentic tasks, complex software engineering, analytics and knowledge work.

Alongside Fable 5, Anthropic is also releasing Claude Mythos 5 to selected partners already approved through Project Glasswing, the company’s controlled-access programme for cybersecurity professionals and other trusted organisations.

The distinction between the two models is central to Anthropic’s strategy. Claude Fable 5 is broadly available, but includes safety classifiers designed to refuse or redirect certain high-risk requests. Claude Mythos 5 shares the same underlying capabilities, but is being made available only in limited settings through Project Glasswing.

The release represents a major shift for Anthropic. Earlier this year, the company had held back Mythos-class models from general release because of concerns around their ability to identify and reason through cybersecurity weaknesses. Anthropic’s concern was that models with such capabilities could help defenders find vulnerabilities, but could also make it easier for malicious actors to discover and exploit flawed software.

That dual-use risk has shaped the entire rollout. Rather than releasing the most powerful version directly to the public, Anthropic is offering Fable 5 as a safeguarded general release, while reserving Mythos 5 for vetted customers.

According to Anthropic’s documentation, Fable 5 and Mythos 5 both became available on June 9, 2026. Fable 5 is available through the Claude API, Claude Platform on AWS, Amazon Bedrock, Vertex AI and Microsoft Foundry. Mythos 5 is not generally available and requires approval through Project Glasswing.

Both models support a one million token context window by default and can produce up to 128,000 output tokens per request. They are priced at $10 per million input tokens and $50 per million output tokens.

Anthropic has also introduced specific refusal and fallback behaviour for Fable 5. When the model declines a restricted request, developers can configure systems to retry with another Claude model. The company says users are not charged for requests refused before output is generated.

The launch follows weeks of scrutiny around Mythos-class capabilities. In April, Anthropic formally introduced the Mythos tier and described it as a major leap in AI capability, particularly for cybersecurity analysis. Rather than making it broadly available, the company created Project Glasswing to give selected cyber defenders early access.

Project Glasswing initially provided access to a small group of partners, including security teams and software maintainers. Anthropic later expanded the initiative to approximately 150 additional organisations across more than 15 countries, prioritising groups responsible for critical infrastructure, open-source software and widely used codebases.

The company has said Project Glasswing partners used Mythos Preview to scan codebases and identify thousands of serious security flaws. That result strengthened Anthropic’s argument that powerful AI models could help defenders move faster, but it also underscored why the company was cautious about broad release.

The arrival of Fable 5 suggests Anthropic believes it has now built sufficient guardrails to offer Mythos-level capability to a wider audience, while still restricting the most sensitive use cases.

Reuters reported that the public model is designed to block risky cybersecurity requests and fall back to a less capable model when necessary. That approach reflects the broader industry challenge: frontier models are becoming more capable across domains where the same skills can be used for legitimate defence or harmful activity.

The timing is also notable. The release comes shortly after Anthropic confidentially filed paperwork for a potential initial public offering in the United States. The filing places the company deeper into the race among major AI firms seeking both technical leadership and financial scale.

For customers, Claude Fable 5 gives broader access to capabilities that had previously been tightly limited. For developers and enterprises, it may strengthen Claude’s position in software engineering, analysis, long-context workflows and agentic systems.

For regulators and cybersecurity experts, however, the release will likely intensify questions about how frontier AI models should be controlled once they cross major capability thresholds.

Anthropic’s strategy appears designed to walk a careful line. The company is trying to make advanced capabilities commercially useful and widely available, while keeping its most sensitive cyber features within a trusted-access framework.

That balance will be closely watched. AI companies are under pressure to keep pace with rivals, satisfy customers and justify soaring valuations. At the same time, governments and security researchers are increasingly concerned that advanced models could lower the barrier for cyberattacks, biological risk, fraud and other harmful activity.

Fable 5 therefore arrives as both a product launch and a test case. It shows how one of the world’s leading AI labs is attempting to release more powerful models without fully abandoning the safety controls that delayed their public availability in the first place.

The release also signals where the industry is heading. The question is no longer only whether AI models can write code, answer complex questions or complete business tasks. The larger question is how companies should release models when their capabilities begin to reshape cybersecurity itself.

Anthropic’s answer, at least for now, is a two-track rollout: Fable 5 for the public, Mythos 5 for trusted partners, and a set of safeguards designed to keep the most dangerous uses out of general circulation.

Whether that balance holds will become one of the most important questions in the next phase of the AI race.

Photo Credit: DepositPhotos.com