Claude Mythos Confirmed March 2026 Confirmed March 2026

Meet Claude Capybara

Anthropic's most powerful AI model ever built. A new tier above Opus with a step change in coding, reasoning, and cybersecurity capabilities.

Fortune Exclusive March 26, 2026 — Anthropic confirmed testing "Claude Mythos" after a CMS leak exposed draft announcements. The new Capybara tier sits above Opus as the most capable Claude ever built.

A new tier in the Claude family

Capybara introduces a fourth tier to Claude's lineup — larger, more intelligent, and more capable than Opus, which was previously Anthropic's most powerful offering.

Tier 1
Haiku
Fast, lightweight, cost-efficient. Built for high-throughput tasks.
Tier 2
Sonnet
Balanced performance. The best price-to-capability ratio.
Tier 3
Opus
Complex reasoning and research-grade analysis. Previously the best.

Dramatically higher scores

Based on Anthropic's leaked internal benchmarks, Capybara dramatically outperforms Claude Opus 4.6 across every major category.

"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."

— Anthropic draft blog post, leaked March 26, 2026 · Source: Fortune

💻 Software Coding

Capybara scores dramatically higher than Opus 4.6 on coding benchmarks. Opus 4.6 already led industry rankings alongside GPT-5.3-Codex.

🧠 Academic Reasoning

Significant improvements in academic reasoning tasks. Anthropic describes Capybara as a "step change" in overall intelligence.

🔒 Cybersecurity

"Currently far ahead of any other AI model in cyber capabilities" — can discover and exploit vulnerabilities faster than human defenders can patch them.

⚡ Overall

Anthropic calls it "the most capable we've built to date" — a new tier, not an incremental version update. Larger architecture than Opus.

All claims sourced from Anthropic's leaked draft blog post and official spokesperson statement.

Capybara vs Opus 4.6

Opus 4.6 scores are from Anthropic's published data. Capybara scores are unreleased — Anthropic confirms "dramatically higher" across the board.

Terminal-Bench 2.0 Agentic Coding

Capybara
?
Opus 4.6
65.4%

Humanity's Last Exam Reasoning

Capybara
?
Opus 4.6
53.1%

Cybersecurity "Far ahead of any other AI"

Capybara
?
Opus 4.6

Finance Agent Enterprise Tasks

Capybara
?
Opus 4.6
60.7%
Capybara (unreleased — "dramatically higher") Opus 4.6 (verified scores)

Opus 4.6 scores sourced from Anthropic's published benchmarks. Capybara bars are illustrative — exact scores pending release. Cybersecurity: no public Opus 4.6 score; Anthropic states Capybara is "far ahead of any other AI model."

Everything confirmed so far

From the Fortune exclusive and Anthropic's official statements.

🏗️

New tier above Opus

"Capybara is a new name for a new tier of model: larger and more intelligent than our Opus models."

"Step change" in capability

Anthropic spokesperson confirmed it is "a step change" in AI performance and "the most capable we've built to date."

🔐

Leading cyber capabilities

"Currently far ahead of any other AI model in cyber capabilities" — can exploit vulnerabilities faster than defenders can patch them.

🧪

Early access testing

Currently being trialed by select early access customers. Anthropic is "being deliberate" about its release given its power.

💰

Premium pricing expected

Described as "expensive to run" and positioned above Opus. Will be Anthropic's most premium model tier.

📅

Public release TBD

No official timeline. The leaked blog had a date, but Anthropic hasn't committed to any schedule post-leak.

Be ready when Capybara launches

Two ways to get ahead of the curve — managed cloud hosting or API access at competitive prices.

Managed Hosting

MyClaw

Deploy your AI stack on managed cloud infrastructure. Capybara access the moment it launches. Zero DevOps, one-click setup.

  • Day-one Capybara model support
  • One-click OpenClaw deployment
  • Dedicated instance, not shared
  • Built-in API proxy & key management
  • Auto-scaling for heavy workloads
  • 24/7 monitoring & updates
Deploy on MyClaw
API Marketplace

Capybara API Access

Buy and sell Claude Capybara API access at market-driven prices. For developers, teams, and enterprises who want the best deal.

  • Competitive Capybara API pricing
  • Buy & sell API keys and token bundles
  • Team & enterprise volume discounts
  • All Claude tiers: Haiku → Capybara
  • Usage analytics & spend tracking
  • DIT.ai waitlist now open
Save up to 40% vs direct
Get API Access

From leak to launch

Tracking every development around Claude Capybara.

February 2026

Opus 4.6 released

Anthropic ships its previous flagship alongside OpenAI's GPT-5.3. Both flagged unprecedented cybersecurity capabilities.

March 26, 2026

The Fortune leak

~3,000 unpublished assets found in Anthropic's unsecured CMS, including the draft blog post announcing Claude Mythos and the Capybara tier.

March 27, 2026

Anthropic confirms

Official statement: "a step change" and "the most capable we've built to date." Early access customers actively testing.

Q2 2026 (Expected)

Limited API rollout

Capybara API expected for approved developers and enterprises, starting with cybersecurity-focused organizations.

2026 (TBD)

General availability

Public API and Claude.ai access will follow Anthropic's rollout. MyClaw will offer managed hosting from day one.

Common questions

Everything you need to know about Claude Capybara and Mythos.

What is Claude Capybara? +
Claude Capybara (internal codename: Mythos) is Anthropic's next-generation AI model. It introduces a new fourth tier in the Claude hierarchy — above Haiku, Sonnet, and Opus. Anthropic describes it as "by far the most powerful AI model we've ever developed," with dramatically higher scores in software coding, academic reasoning, and cybersecurity benchmarks compared to Claude Opus 4.6.
What's the difference between Capybara and Mythos? +
They refer to the same underlying model. "Claude Mythos" is the product/generation name (like "Claude 4"), while "Capybara" is the tier name (like Haiku, Sonnet, Opus). The full designation would be something like "Claude Mythos Capybara" — the Capybara-tier model from the Mythos generation.
When will Capybara be publicly available? +
No official release date has been announced. As of March 2026, the model is in early access testing with select customers. Anthropic describes it as "expensive to run and not yet ready for general release." Given their cautious approach, expect a phased rollout starting with enterprise and cybersecurity customers.
How much will Claude Capybara cost? +
Official pricing hasn't been released. As a tier above Opus (currently $15/$75 per million input/output tokens), Capybara will likely be Anthropic's most premium model when it launches.
How can I get Capybara as soon as it launches? +
MyClaw — our managed hosting platform will integrate Capybara on day one, so you can start using it from your own cloud instance immediately.
Is Capybara better than GPT-5.3? +
Based on Anthropic's internal assessments, Capybara is "currently far ahead of any other AI model in cyber capabilities" and dramatically outperforms Opus 4.6 in coding and reasoning. Since Opus 4.6 was already competitive with GPT-5.3-Codex, Capybara would represent a significant leap beyond the current frontier from any lab.
What makes Capybara different from Opus? +
Capybara isn't an incremental upgrade — it's a "step change." It's a new, larger model tier, not a version bump. Think of the gap between Sonnet and Opus, applied above Opus. Key differences: dramatically better coding, breakthrough cybersecurity capabilities, enhanced reasoning, and what Anthropic considers the threshold for AI that can "exploit vulnerabilities faster than defenders can patch them."
Why was Capybara leaked? +
A CMS misconfiguration at Anthropic left ~3,000 unpublished assets in a publicly accessible, searchable data store. Fortune discovered and reported on it March 26, 2026. Security researchers at Cambridge and LayerX independently verified the materials. Anthropic attributed the leak to "human error in the CMS configuration."

Don't miss the Capybara wave

Join thousands of developers and teams preparing for the most powerful Claude ever built.

Not affiliated with Anthropic. Claude, Capybara, Mythos, Opus, Sonnet, and Haiku are trademarks of Anthropic, PBC. Last updated March 27, 2026.