Anthropic’s strongest model revealed, achieves a “generational leap” in performance, thoroughly surpassing Opus!

```

A misconfiguration within Anthropic accidentally exposed its most confidential technological trump card to the public.

According to an exclusive report by Fortune magazine on the 26th, a brand new flagship model named "Claude Mythos", internally codenamed "Capybara", has completed training and entered early testing. Anthropic officially describes its performance improvements as a "generational leap", and in core tests such as programming, academic reasoning, and cybersecurity, it thoroughly surpasses the current strongest version, Claude Opus 4.6.

The report states that due to a human error in Anthropic's content management system (CMS), nearly 3,000 unpublished internal documents were exposed in a publicly searchable data cache. The leaked documents reveal the company's plan to introduce a brand new model tier named "Capybara", positioned higher than the existing flagship Opus series, with even higher operating costs.

However, the superior performance brings unprecedented safety risks. The leaked internal draft shows that Anthropic characterizes the new model as a major source of security risk in the field of cybersecurity, stating that its cyberattack capabilities "far surpass any other AI model available today". Once exploited by hackers, it could give rise to massive cyberattacks with destructive power surpassing defenders' ability to respond—this is the core reason the company has hesitated to release it publicly.

Surpassing Opus: The new "Capybara" tier breaks existing patterns

The leaked blog draft shows that Claude Mythos represents a structural overhaul of Anthropic’s product line.

Currently, Anthropic’s model matrix includes three tiers: the most powerful Opus, the balanced Sonnet for speed and cost, and the smallest, fastest Haiku. The leaked documents show Anthropic is introducing a new "Capybara" tier positioned above Opus—larger in scale, more intelligent, but also more expensive to run.

In specific performance areas, the draft blog says: "Compared to the previous strongest version Claude Opus 4.6, Capybara’s scores in software programming, academic reasoning, and cybersecurity tests have all risen sharply." The documents describe Claude Mythos as "the most powerful AI model we have ever developed, far surpassing any previous version".

In response to Fortune’s inquiry, an Anthropic spokesperson confirmed the above direction, stating the company is "developing a general-purpose model with major advances in reasoning, programming, and cybersecurity," and emphasizing that "given its powerful capabilities, we are cautiously advancing the release strategy".

Unprecedented Cybersecurity Risks: Anthropic itself is wary

Accompanying these powerful capabilities are unusually strong security warnings in the leaked documents.

The leaked blog draft bluntly states that the new model "currently far exceeds any other AI model in terms of cyberattack capabilities," and presages "an upcoming wave of models whose vulnerability exploitation speed will far outpace defenders’ response capabilities."

Because of these concerns, Anthropic's release strategy prioritizes cybersecurity defense agencies. The draft writes: "We will first grant early access to relevant agencies so they can gain an edge in the upcoming wave of AI-powered vulnerability exploitation, strengthening the robustness of their codebases in advance."

This concern is not isolated. In February this year, OpenAI released GPT-5.3-Codex, the first model rated "high capability" in cybersecurity tasks, and also the first to be directly trained to identify software vulnerabilities. At the same time, Anthropic’s Claude Opus 4.6 also showed the ability to identify unknown vulnerabilities in production codebases, which the company acknowledged as a "double-edged sword" with both offensive and defensive uses.

Misconfiguration: 3,000 internal files accidently made public

The technical root of this leak incident lies in what seems like a low-level operational mistake.

Roy Paz, a senior AI security researcher at LayerX Security, and Alexandre Pauwels, a cybersecurity researcher at Cambridge University, discovered that an external CMS tool used by Anthropic for publishing official blogs had a configuration flaw: the system by default set uploaded digital assets to public and assigned accessible URLs unless users manually set them as private. This default caused nearly 3,000 unpublished assets—including images, PDFs, and audio files—to be exposed in a publicly searchable data cache.

In its statement, Anthropic attributed the incident to "human error", saying "an issue with the external CMS tool resulted in draft content being publicly accessible", and described the leaked materials as "early drafts of content considered for publication".

X platform erupts: Shock and doubts coexist

As soon as the news was exposed, X platform erupted in discussion, with public attention focused on both technological shock and security trust.

Account TFTC directly pointed out the irony of the event: "Anthropic accidentally proves the case for AI safety. A CMS misconfiguration resulted in 3,000 unpublished files exposed in a publicly searchable cache—including details about 'Claude Mythos' (codename Capybara), which Anthropic's internal papers call 'the most powerful AI to date'." This comment spread widely in the tech community, and many users expressed shock that a company whose core value is AI safety exposed its secrets by such a basic misconfiguration.

Another user, fardeen, turned attention to Claude's latest abilities, commenting: "Claude can now really use a computer like you—open apps, click buttons, fill out forms. Anthropic is gradually removing humans from the operation flow."

User Oliwier Makowski Trusz believes the release of Capybara will change the landscape. According to leaked info, its parameter count reaches 10 trillion. The gap between Claude and all other models has just widened significantly.

Risk warning and disclaimerThe market has risks, investment needs caution. This article does not constitute personal investment advice, nor does it take into account individual users' special investment goals, financial situations, or needs. Users should consider whether any opinions, views, or conclusions in this article suit their particular circumstances. Invest accordingly at your own risk. ```