Leak Watch

Claude Mythos
A Possible Tier Above Opus

Based on late-March leaked draft materials and public reporting, Capybara may be Anthropic's next model tier, with Claude Mythos as the first surfaced model name.

#ClaudeMythos #Capybara #Anthropic #LeakWatch

What Are Capybara and Claude Mythos

Current public knowledge comes mainly from Fortune reporting, the CMS exposure discovered by LayerX / Cambridge researchers, and follow-up media coverage. The overall picture suggests Capybara is a possible new tier above Haiku, Sonnet, and Opus, while Claude Mythos is the first leaked model name within that tier.

Leaked quote
"the most capable we've built to date"
Tier 1
Haiku
Haiku — Fast and low-cost
Tier 2
Sonnet
Sonnet — Balanced default
Tier 3
Opus
Opus — Current flagship
Tier 4?
Capybara
Capybara — Higher leaked tier

What The Leak Claims

Stronger coding

The leaked draft says its software coding performance is materially higher than Claude Opus 4.6.

Deeper reasoning

Public descriptions point to clear gains in academic reasoning and complex knowledge synthesis.

Cybersecurity focus

The draft repeatedly emphasizes cyber offense and defense capability as the main reason for a cautious rollout.

Higher-tier positioning

If Capybara launches as a real tier, Anthropic's product ladder would extend beyond Opus for the first time.

Relative Positioning vs Released Models

Dimension Claude Mythos / Capybara Claude Opus 4.6 GPT-5.4
Coding Leak says clearly above Opus 4.6 Anthropic's current released flagship Top competing model
Reasoning Leak says meaningful step-up Current strength Current strength
Cybersecurity Described as the highest-risk capability jump Already strong Industry capabilities also rising
Context / spec Not publicly confirmed 1M context available Await official product details
Pricing Rumored to be above Opus, unconfirmed Officially published Depends on official product tier
Status Early testing / unreleased Released Released

This page is based on public-source synthesis and should not be treated as Anthropic's final naming, pricing, API availability, or capability commitment.

Public Timeline

The sequence below is based on public reports, leaked draft materials, and cross-checkable sources.

01

March 26-27, 2026: An Anthropic CMS configuration mistake exposed unpublished draft assets, and Fortune first reported the Mythos / Capybara references.

02

Current status: Anthropic has confirmed it is testing a stronger general-purpose model with a small set of early-access customers.

03

Possible rollout path: the leaked draft suggests a cautious release focused first on security defenders and enterprise-style evaluation.

04

Positioning takeaway: existing signals look more like a new tier above Opus than a routine minor-version refresh.

This page is based on public-source synthesis and should not be treated as Anthropic's final naming, pricing, API availability, or capability commitment.

What This Means For Developers

The coding ceiling keeps rising

If Mythos ships, complex refactors, long-horizon agent coding, and high-stakes code review workflows could improve again.

Security becomes the main battleground

Anthropic's strongest caution is not chat quality but cyber capability, which matters directly for red-teaming, auditing, and vuln remediation.

Enterprise pilots matter more

High cost, high capability, and cautious release language all point to an enterprise-first evaluation pattern.

Preparing early pays off

Teams already organized around Claude Code, Opus 4.6, and long-context workflows will be able to adopt stronger models faster later.

Further Reading

Get familiar with the currently released Claude models, developer APIs, and QCode.cc workflows first.

Use the best Claude available today, watch Mythos next

QCode.cc already supports today's main Claude models and developer workflows. If a new model launches officially, it will fit more naturally into your existing setup.