What Is Claude Mythos (Capybara)? The New AI Tier Above Opus

TL;DR: Claude Mythos is Anthropic’s newest and most powerful AI model. It belongs to a brand-new tier called Capybara, which sits above the existing Opus tier. According to leaked draft documents, it delivers dramatically higher scores in coding, academic reasoning, and cybersecurity. The name “Mythos” was chosen to evoke deep, foundational knowledge that connects ideas across domains.

Claude Mythos and Capybara: What’s the Difference?

Capybara and Mythos are not the same thing, even though the two names are often used interchangeably. The distinction matters.

Capybara is a tier name — a product category in Anthropic’s model lineup, just like Haiku, Sonnet, and Opus. It represents an entirely new performance class that sits above everything Anthropic has released before.

Mythos is the first model released within the Capybara tier. Think of the relationship the same way “Claude 3.5” is a model within the Sonnet tier. There may eventually be other Capybara-tier models, but Mythos is the inaugural one.

This is not an incremental Opus upgrade. According to the leaked draft blog, Capybara represents a fundamentally new architectural level. Anthropic is not simply scaling Opus up — they are introducing a new tier with different capabilities and design goals.

Where Claude Mythos Fits in Anthropic’s Model Family

Anthropic organizes its Claude models into tiers, each targeting different use cases and performance levels. Here is how the full hierarchy looks now that Capybara has entered the picture:

Haiku is the lightweight tier. It is optimized for speed and cost, delivering sub-second responses for tasks like classification, summarization, and simple Q&A. Haiku is what you reach for when latency matters more than depth.

Sonnet is the balanced workhorse. It handles most everyday tasks — writing, analysis, moderate coding — with a strong mix of quality and speed. For the majority of users, Sonnet is the default choice.

Opus is the flagship reasoning tier. It excels at complex multi-step problems, advanced coding, nuanced writing, and tasks that demand deep thinking. Until now, Opus was the ceiling of what Claude could do.

Capybara (Mythos) is the new tier beyond Opus. According to leaked documents, it is both larger and more intelligent than any Opus model. It targets the hardest problems in software engineering, academic research, and cybersecurity — domains where even Opus hits its limits.

The introduction of Capybara does not replace or deprecate Opus. Each tier continues to serve its niche. Capybara simply extends the top end of what Anthropic offers.

What the Leaked Documents Say About Claude Mythos

Several draft blog posts and internal documents have surfaced that describe Claude Mythos in detail. While Anthropic has not officially published these materials, the language is specific enough to paint a clear picture.

According to the leaked draft blog, Anthropic describes the Capybara tier in unambiguous terms: “Capybara is a new name for a new tier of model: larger and more intelligent than our Opus models.” This confirms that Capybara is not a rebrand — it is a step above.

The documents also claim “dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity.” No specific benchmark numbers have been published, but the framing suggests a meaningful gap over Opus rather than a marginal improvement.

Perhaps most notably, the draft refers to Mythos as “by far the most powerful AI model we’ve ever developed.” That language is unusually strong for Anthropic, a company that tends toward measured claims. It signals genuine confidence in a significant capability jump.

The leaked materials also mention early access for cyber defenders, suggesting Anthropic sees Mythos as particularly relevant to security applications where advanced reasoning can identify vulnerabilities and threats that simpler models miss.

Why Anthropic Chose the Name ‘Mythos’

Anthropic’s naming choices have always carried meaning. Haiku evokes brevity. Sonnet suggests structured elegance. Opus implies a grand, complex work. Mythos continues this tradition.

According to the leaked draft blog, Anthropic chose the name Mythos to “evoke the deep connective tissue that links together knowledge and ideas.” The word comes from the Greek (mythos), meaning a foundational narrative — the kind of story that shapes an entire culture’s understanding of reality.

The name suggests that Anthropic sees this model as something more than a reasoning engine. Mythos is meant to connect disparate domains of knowledge in the way that foundational narratives connect disparate aspects of human experience. It is a model that does not just answer questions but synthesizes understanding.

Whether the name lives up to the ambition remains to be seen. But the choice signals that Anthropic views Capybara-tier models as qualitatively different from what came before — not just faster or more accurate, but deeper in how they process and relate information.

What We Know vs What We Don’t About Claude Mythos

Given the mix of leaked documents and official silence, it is worth being precise about what is confirmed and what remains speculation.

Confirmed:

• The model exists. Multiple leaked documents describe it in detail, and Anthropic has not denied its existence.
• Capybara is a new tier, distinct from and above Opus. This is stated explicitly in the draft blog materials.
• Early access is being provided to cyber defenders. The leaked documents specifically mention this as part of the rollout strategy.
• The name “Mythos” was chosen deliberately to reflect the model’s approach to connecting knowledge.

Unconfirmed:

• ⚠️ Specific benchmark scores. The documents say “dramatically higher” but no numbers have surfaced.
• ⚠️ Pricing. There is no leaked or official information about what Capybara-tier access will cost.
• ⚠️ Context window size. No details on whether Mythos supports a larger context window than Opus.
• ⚠️ General availability date. Early access for security researchers is confirmed, but a public release timeline has not been shared.
• ⚠️ Architecture details. The documents describe it as “larger” but do not specify parameter counts, training approaches, or architectural innovations.

Treat any claims beyond the confirmed items with appropriate skepticism. The leaked documents are drafts, and details may change before any official announcement.

Further Reading

• Claude Model Comparison — See how Mythos stacks up against Haiku, Sonnet, and Opus across key dimensions.
• Security Impact of Claude Mythos — A deeper look at what Capybara-tier models mean for cybersecurity.
• The Claude Mythos Leak Explained — Full breakdown of what the leaked documents contain and where they came from.