A new, more capable Claude model appears to have finished training at Anthropic, according to unconfirmed reports circulating on social media, though the company has not officially announced it. The development comes just 10 days after the U.S. government ordered Anthropic to disable its most powerful model, Claude Fable 5, citing national security concerns over a potential jailbreak vulnerability.

What Actually Happened to Fable 5?

On June 9, Anthropic released two frontier-class models at the top of its lineup: Claude Fable 5 and Claude Mythos 5. These models represent a new tier called Mythos-class, positioned above Anthropic's previous top tier, Opus. Both models are actually the same underlying system, but with different safety settings. Fable 5 includes guardrails that prevent it from answering sensitive questions about cybersecurity, biology, and chemistry. Mythos 5 removes those guardrails entirely and is restricted to vetted partners through a program called Project Glasswing.

The public-facing Fable 5 came with impressive specifications: a context window of 1 million tokens, meaning it could process roughly 100,000 words at once, with a maximum output of 128,000 tokens, priced at $10 per million input tokens and $50 per million output tokens. The model was described as state-of-the-art on nearly every benchmark, with particular strength in software engineering, knowledge work, vision, and science tasks.

Three days after launch, on June 12 at 5:21 p.m. ET, everything changed. The U.S. government issued an export-control directive ordering Anthropic to suspend all access to both Fable 5 and Mythos 5 for any foreign national, anywhere in the world, including Anthropic's own foreign-born staff. The stated reason was national security; the government believed someone had discovered a way to jailbreak the model. Anthropic's own review found the technique was essentially asking the model to read a codebase and fix its flaws, something the company argued is already widely available from other model providers. Rather than attempt a partial compliance, Anthropic disabled both models entirely for all users.

What's the New Model Rumor, and Is It Real?

On June 22, unconfirmed reports began circulating that a new, more capable Mythos-class model had finished training. Additionally, a slug labeled "claude-sonnet-5" appeared on an Anthropic partner provider, suggesting either a new Sonnet-tier model or a different naming convention for the next generation. These claims come exclusively from social media posts and have not been confirmed by Anthropic or listed in the company's official model documentation.

The distinction matters because Anthropic's model lineup includes multiple tiers. Haiku is the fastest and most compact; Sonnet balances speed and capability for everyday tasks; Opus sits at the top of the previous tier; and Mythos-class now represents the frontier tier above Opus. A new Mythos-class model would be more significant than a Sonnet update, but the concrete evidence so far is limited to a single slug sighting, which is not confirmation of a full release.

How to Interpret the Signals and Separate Fact From Speculation

• Confirmed Facts: Fable 5 and Mythos 5 launched on June 9; they are the same model with different safety settings; the U.S. government issued an export order on June 12; Anthropic disabled both models for all users; no new model appears in Anthropic's official documentation as of June 22.
• Unconfirmed Signals: A new, more capable Mythos-class model has finished training according to social media reports; a "claude-sonnet-5" slug has been spotted on a partner provider; these claims have not been verified by Anthropic or major news outlets.
• Speculation: Some observers theorize that the export embargo, by cutting off access for foreign users, freed up computing resources that Anthropic could redirect toward training a new model; this is plausible but unconfirmed.

The most likely near-term release is Claude Sonnet 5, since the concrete slug sighting points to the Sonnet tier, which is a workhorse model that would draw less regulatory scrutiny than another frontier-class release. A new Mythos-class model, if it exists, may remain internal or limited to vetted partners for now, given the regulatory pressure surrounding Fable 5.

Why This Matters Beyond the Model Names

The real lesson from the Fable 5 shutdown extends beyond speculation about what's next. On June 9, Fable 5 was the most powerful language model available to the public. Three days later, it was gone for most of the planet, not because the model degraded, but because of a force entirely outside Anthropic's control. For developers and organizations that had begun building tools on top of Fable 5, the sudden unavailability created a problem: their applications stopped working overnight.

However, the impact was not uniform. Organizations that built on top of Anthropic's broader system, rather than a single model, were able to swap to the next-best available Claude model with minimal friction. This highlights a critical architectural lesson for AI adoption: relying on a specific frontier model carries regulatory and availability risk, while building on a flexible system that can swap models maintains resilience.

As of now, all other Claude models remain fully online and available. Anthropic has not announced any timeline for when Fable 5 or Mythos 5 will return, or under what conditions they might be re-enabled. The company's position is that perfect jailbreak-resistance is not achievable for any model provider, and if that were the standard, no new frontier model could ever launch.

The unconfirmed reports of a new model suggest Anthropic is moving forward with development despite the regulatory headwinds. Whether that model launches publicly, remains internal, or follows a restricted-access model like Project Glasswing will likely depend on how the company addresses the government's security concerns. Until Anthropic makes an official announcement, the rumors remain exactly that: interesting signals from credible observers, but not confirmed fact.