The AI industry is no stranger to high-stakes surprises — but few events in 2026 have shaken the tech world quite like the accidental unveiling of Claude Capybara. On March 26–27, 2026, a configuration error inside Anthropic’s content management system exposed nearly 3,000 unpublished internal files to the open internet. Hidden inside that data trove was a draft blog post detailing what Anthropic internally describes as “by far the most powerful AI model we’ve ever developed.” The model carries two names used interchangeably in leaked documents: Claude Mythos and Claude Capybara.
This was not a planned announcement. Security researchers from LayerX Security and the University of Cambridge discovered the exposed assets, Fortune reviewed the documents, and the rest of the internet quickly took notice. For developers, enterprise AI teams, and cybersecurity professionals, this leak raises urgent questions: What exactly is Claude Capybara? How does it stack up against the current flagship, Claude Opus 4.6? And when — if ever — can you actually use it?
This article breaks down everything confirmed, everything credibly leaked, and what it all means for the future of AI.
What Is Claude Capybara (Mythos)?
Claude Capybara is the internal name for Anthropic’s next-generation AI model, also referred to as Claude Mythos in leaked draft documents. Anthropic has described it as “a step change” in AI performance and “the most capable we’ve built to date,” positioned as a new tier surpassing their previous flagship Opus models in size, intelligence, and capability.
This is not a routine model update. Capybara introduces a fourth tier to Claude’s model lineup architecture — larger, more intelligent, and more capable than Opus, which was previously Anthropic’s most powerful offering. Think of it less as “Opus 5” and more as an entirely new product category.
According to the leaked draft, the name “Mythos” was chosen to evoke “the deep connective tissue that links together knowledge and ideas.” Whether the model ultimately ships under the Mythos or Capybara label — or something else entirely — remains unconfirmed as of April 2026.
How the Claude Capybara (Mythos) Leak Happened
The story behind the Claude Capybara (Mythos) leak is itself a cautionary tale about operational security at frontier AI labs.
A draft blog post that was available in an unsecured and publicly searchable data store prior to Thursday evening said the new model is called Claude Mythos. The AI lab left the material, including what appeared to be a draft blog post announcing a new model, in an unsecured, public data lake, according to documents separately located and reviewed by Roy Paz, a senior AI security researcher at LayerX Security, and Alexandre Pauwels, a cybersecurity researcher at the University of Cambridge.
Digital assets, including images, PDF files, and audio files, were set to public by default upon upload in the CMS, unless explicitly marked private. This oversight led to approximately 3,000 assets linked to Anthropic’s blog becoming publicly accessible.
Anthropic attributed the incident to “human error” and described the exposed material as early drafts of content being considered for publication. A second, separate leak followed just days later: Anthropic accidentally published all of Claude Code’s original source code to NPM, a platform developers use to share software, instead of only the finished compiled version. The source code leak exposed around 500,000 lines of code across roughly 1,900 files — and provided additional corroboration that the Capybara model was actively being prepared for launch.
The irony? A company warning that its new model poses unprecedented cybersecurity risks had itself committed a basic operational security error.
Capybara vs. Mythos: Are They the Same Model?
One of the most common points of confusion surrounding this story is whether “Capybara” and “Mythos” refer to different models. They do not.
“Claude Mythos” is the product or generation name — similar to how “Claude 4” functions — while “Capybara” is the tier name, analogous to Haiku, Sonnet, or Opus. The full designation would be something like “Claude Mythos Capybara.”
There were two versions of the same blog post that only differed in the model’s name: “Mythos” (v1) and “Capybara” (v2). In the Capybara version, the name was swapped throughout the title and body text, but the subtitle still reads, “We have finished training a new AI model: Claude Mythos.” Anthropic told Fortune the documents were simply early drafts, suggesting the company had not yet finalized which name to use publicly.
Claude’s New Model Tier Structure Explained
To understand why Claude Capybara matters, you need to understand how Anthropic structures its model lineup.
Anthropic currently offers models in three tiers: Opus (most capable), Sonnet (faster and cheaper), and Haiku (smallest and fastest). Capybara would add a fourth, pricier tier above all three.
Here is a simplified breakdown of the current and upcoming hierarchy:
| Tier | Speed | Cost | Capability |
|---|---|---|---|
| Haiku | Fastest | Cheapest | Basic tasks |
| Sonnet | Fast | Moderate | Balanced performance |
| Opus | Moderate | Expensive | Research-grade reasoning |
| Capybara | TBD | Most Expensive | Step-change capabilities |
The Capybara tier is described not as an incremental improvement but as a structural expansion of what Anthropic’s models can do — particularly in software engineering, academic reasoning, and cybersecurity.
Opus 4.6 vs. Claude Capybara: How Do They Compare?
This is the comparison every developer and enterprise AI buyer is asking about. Based on the leaked documents, the performance gap between Opus 4.6 vs. Claude Capybara is significant.
According to the leaked draft, “Compared to our previous best model, Claude Opus 4.6, Capybara gets dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity, among others.”
Opus 4.6 had only recently topped Terminal-Bench 2.0 at 65.4%, surpassing GPT-5.2-Codex. If Capybara scores “dramatically higher” on the same evaluations, the performance ceiling for AI-assisted coding and analysis could be meaningfully raised.
Key differentiators based on leaked claims:
- Software coding: Capybara significantly outperforms Opus 4.6 on complex multi-file engineering tasks.
- Academic reasoning: Better multi-step logic, research synthesis, and mathematical reasoning.
- Cybersecurity: The most notable gap — Capybara is described as “currently far ahead of any other AI model in cyber capabilities.”
- Cost: The draft acknowledged the model is “very expensive for us to serve, and will be very expensive for our customers to use.”
It’s worth noting that no official benchmark numbers have been publicly released by Anthropic as of April 2026. All performance claims originate from the leaked draft and should be treated as preliminary until verified.
Cybersecurity Capabilities and Dual-Use Concerns
The single most consequential aspect of Claude Mythos is not its reasoning ability or coding prowess — it’s the dual-use cybersecurity implications.
Anthropic appears to be especially worried about the model’s cybersecurity implications, noting that the system is “currently far ahead of any other AI model in cyber capabilities,” and that “it presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders.”
Anthropic’s own leaked draft stated plainly: “In preparing to release Claude Capybara, we want to act with extra caution and understand the risks it poses — even beyond what we learn in our own testing.” This is not boilerplate AI safety language. The concern is grounded in documented real-world incidents.
Anthropic discovered that a Chinese state-sponsored group had already been running a coordinated campaign using Claude Code to infiltrate roughly 30 organizations — including tech companies, financial institutions, and government agencies — before the company detected it.
Shares of cybersecurity vendors including CrowdStrike, Palo Alto Networks, Zscaler, and Fortinet fell as investors assessed what more capable models within Claude Code Security could mean for the competitive landscape.
The planned rollout strategy reflects this caution. Anthropic wrote in the draft blog post that it would slowly expand access to Claude Mythos to more customers using the Claude API over the coming weeks, with an initial focus on cybersecurity use cases through an early access program.
How to Use Claude Capybara (Mythos) — Access & Availability
If you’re wondering how to use Claude Capybara (Mythos), the answer right now is: you almost certainly cannot.
As of April 2026, Claude Mythos / Capybara is available only to a small group of early access customers selected by Anthropic. There is no public API, no announced pricing, and no confirmed release date.
The leaked rollout strategy outlines a deliberately phased approach, quite different from how Anthropic has released previous models:
- Phase 1: Small group of early-access enterprise customers, focused on cybersecurity defense applications.
- Phase 2: Gradual expansion via the Claude API to vetted customers.
- Phase 3: Broader release — contingent on efficiency improvements and safety evaluations.
The drafts also acknowledge the model is “very expensive for us to serve, and will be very expensive for our customers to use.” Anthropic says it’s working to make it “much more efficient before any general release.”
For developers preparing to integrate Capybara when it does arrive, the transition will be straightforward — switching models in the Claude API requires only a single parameter change. In the meantime, Claude Opus 4.6 and Sonnet 4.6 remain the best options for production workloads.
Market Reaction and Industry Impact
The Claude Capybara (Mythos) leak did not stay confined to AI forums. It moved markets.
The iShares Expanded Tech-Software Sector ETF (IGV) fell nearly 3% in early trading following the leak, likely contributing to bitcoin’s tumble back to $66,000. The reaction underscored how seriously investors are now treating frontier AI model announcements — even accidental ones.
Bloomberg and The Information reported that Anthropic was considering an IPO as early as October 2026, with The Information’s briefing directly mentioning: “Anthropic Discusses Q4 IPO, Preps Advanced ‘Claude Mythos’ / Capybara AI.” If accurate, the timing of Capybara’s general release may be strategically tied to the company’s broader financial roadmap.
Industry analysts remain cautiously optimistic. Avasant’s research director, Gaurav Dewan, noted that “powerful models will not replace cybersecurity platforms,” seeing vendors increasingly embedding frontier models from Anthropic and OpenAI rather than being displaced by them.
FAQ
Q1: What is Claude Capybara?
Claude Capybara is the tier name for Anthropic’s next-generation AI model, internally called Claude Mythos. It sits above the existing Opus tier and is described by Anthropic as the most capable model the company has ever built, with dramatically higher scores in coding, reasoning, and cybersecurity.
Q2: Is Claude Capybara the same as Claude Mythos?
Yes. “Mythos” refers to the product or generation name, while “Capybara” is the tier name — similar to how Haiku, Sonnet, and Opus describe capability levels. The two names appear in different drafts of the same announcement and refer to the same underlying model.
Q3: How did the Claude Capybara (Mythos) leak happen?
A misconfiguration in Anthropic’s content management system left approximately 3,000 unpublished internal assets publicly accessible, including a draft blog post announcing the model. Security researchers at LayerX Security and the University of Cambridge discovered and reported the exposure.
Q4: How does Opus 4.6 compare to Claude Capybara?
Based on leaked documents, Capybara scores “dramatically higher” than Claude Opus 4.6 on software coding, academic reasoning, and cybersecurity benchmarks. Opus 4.6 had previously topped Terminal-Bench 2.0 at 65.4%, making the claimed performance gap significant. Official numbers have not yet been published.
Q5: Can I use Claude Capybara (Mythos) right now?
No. As of April 2026, the model is available only to a small group of early-access enterprise customers. There is no public API access, no pricing announcement, and no confirmed general release date.
Q6: Why is Anthropic being so cautious about releasing Claude Capybara?
Primarily due to cybersecurity concerns. Leaked documents state that Capybara is “currently far ahead of any other AI model in cyber capabilities” and could enable vulnerability exploitation at a scale that outpaces existing defenses. Anthropic is conducting additional safety evaluations before any broader release.
Q7: When will Claude Capybara be publicly available?
No official date has been confirmed. Some analysts speculate the release could align with a potential Anthropic IPO in Q4 2026, but this is speculative. Anthropic has indicated the model needs cost and efficiency improvements before a general rollout.
Conclusion
Claude Capybara — whether it ultimately ships under that name, the Mythos label, or something else entirely — represents a genuine inflection point in Anthropic’s model roadmap. The leaked documents, confirmed by Anthropic itself, paint a picture of a frontier model that is meaningfully more capable than anything the company has previously released, with particularly striking advances in software engineering and cybersecurity.
The irony of a company warning about unprecedented AI-powered cybersecurity risks accidentally leaking its own crown jewel through a CMS misconfiguration will not be lost on anyone in the industry. But beyond the drama, the substance matters: a new fourth tier above Opus is coming, access will be restricted and expensive, and the dual-use implications for cybersecurity are real enough to move financial markets.
For developers and enterprise teams, the actionable takeaway is simple: build on Claude Opus 4.6 today, stay informed about Capybara’s phased rollout, and prepare your API integrations for a straightforward model switch when access eventually opens. Claude Capybara isn’t here yet — but understanding what it is puts you ahead of the curve.
Want to stay ahead of the AI model landscape as it evolves? Explore our in-depth guides on the Claude API, how to benchmark large language models for enterprise use cases, and the latest developments from Anthropic and its competitors. Subscribe to our newsletter for real-time coverage of frontier AI releases — including Claude Capybara the moment public access becomes available.
See More Awesome Post Like this on Our Blog.
