What Is Claude Mythos? Anthropic’s Most Powerful AI Model Explained
Claude Mythos is a new AI model developed by Anthropic that sits in a previously unannounced tier called Capybara — above the existing Opus, Sonnet, and Haiku lineup. Claude Mythos was accidentally revealed on March 27, 2026 when a misconfigured content management system exposed approximately 3,000 internal documents, including draft blog posts describing the model as “by far the most powerful AI model we’ve ever developed.”
Anthropic confirmed the model exists and is being tested with a small group of early access customers. According to leaked materials, Mythos achieves dramatically higher scores than Claude Opus 4.6 on tests of software coding, academic reasoning, and cybersecurity.
- Capybara is a new fourth tier above Opus in Anthropic’s model hierarchy
- Mythos excels at coding, reasoning, and cybersecurity tasks
- No public release date has been confirmed
- Early access is currently limited to defensive cybersecurity organizations
- The leak caused cybersecurity stocks to drop significantly
What Claude Mythos Can Do
Anthropic describes Claude Mythos as a general-purpose model with meaningful advances in three core areas. The first is software coding — Mythos handles complex multi-file codebases, debugging, and code generation at a level that significantly exceeds Claude Opus 4.6. The second is academic reasoning, where the model demonstrates stronger multi-step logical analysis and problem-solving.
The third and most discussed capability is cybersecurity. Leaked documents describe Mythos as “currently far ahead of any other AI model in cyber capabilities.” The model can identify and exploit software vulnerabilities at a speed and accuracy that Anthropic itself considers unprecedented. This dual-use nature — equally powerful for defense and offense — is the primary reason behind Anthropic’s cautious release strategy.
Benchmarks and Performance
Direct benchmark numbers have not been published, but the leaked draft blog post claimed Mythos scores “dramatically higher” than Claude Opus 4.6 across all tested domains. For context, Opus 4.6 already topped Terminal-Bench 2.0 at 65.4%, surpassing OpenAI’s GPT-5.2-Codex. If the leaked claims hold, Mythos would represent the strongest performing AI model currently in existence.
The Capybara Tier: How Anthropic’s Model Hierarchy Works
Anthropic organizes its AI models into tiers based on capability and cost. Until Mythos, the lineup consisted of three tiers. Capybara adds a fourth.
| Tier | Speed | Cost | Capability | Current Model |
|---|---|---|---|---|
| Haiku | Fastest | Lowest | Good | Claude Haiku 4.5 |
| Sonnet | Fast | Medium | Strong | Claude Sonnet 4.6 |
| Opus | Moderate | High | Very Strong | Claude Opus 4.6 |
| Capybara | Slowest | Highest | Breakthrough | Claude Mythos |
The naming follows a pattern: Capybara is the tier, Mythos is the specific model — just as Opus is a tier and Opus 4.6 is a specific release. According to the leaked draft, Capybara is described as “larger and more intelligent than our Opus models — which were, until now, our most powerful.”
The name Mythos was chosen to represent “the deep connective tissue that links knowledge and ideas together.” Anthropic has not confirmed whether this name will appear in final marketing materials or if it will be changed before public release.
How Claude Mythos Was Leaked
The discovery of Claude Mythos was not a planned announcement. Security researchers Roy Paz from LayerX Security and Alexandre Pauwels from the University of Cambridge found approximately 3,000 unpublished assets in a publicly accessible, unencrypted data store. The files included draft blog posts, images, PDFs, and internal documents that were never meant to be public.
The root cause was a misconfiguration in Anthropic’s content management system. Digital assets created via the CMS were set to public by default and assigned publicly accessible URLs upon upload. Someone at Anthropic failed to change the privacy settings, making the entire cache searchable.
Fortune was the first outlet to report the leak after notifying Anthropic on Thursday evening. The company immediately restricted access to the data store and attributed the incident to “human error” in CMS configuration.
What Else Was Exposed
Beyond the Mythos draft, the leaked documents included details about an invite-only CEO retreat in England with Anthropic CEO Dario Amodei attending. The two-day event at an 18th-century manor was described as an “intimate gathering” for European business leaders to discuss AI adoption and experience unreleased Claude capabilities. Other leaked files appeared to include discarded blog assets, unused banners, and at least one document referencing an employee’s parental leave.
The irony was not lost on observers: an AI safety company — known for its cautious approach to powerful AI systems — accidentally exposed its most sensitive project through a basic security misconfiguration.
Cybersecurity Risks and Market Impact
The cybersecurity dimension of Claude Mythos is what separates it from a typical model upgrade. Anthropic’s own leaked documents warned that the model “presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders.”
This is a dual-use problem. The same capability that helps security teams find and patch vulnerabilities can be used by attackers to discover and exploit them. Anthropic acknowledged the risk directly, stating it wants to “act with extra caution and understand the risks it poses — even beyond what we learn in our own testing.”
Stock Market Reaction
The leak triggered immediate selling in cybersecurity stocks. CrowdStrike dropped approximately 7%, Palo Alto Networks fell around 6%, and Fortinet declined 4–6%. The iShares Expanded Tech-Software ETF (IGV) also moved lower. Investors interpreted the news as a signal that AI models capable of finding zero-day vulnerabilities could fundamentally change the cybersecurity landscape, potentially undermining the effectiveness of existing security products.
Past Incidents With Claude
The cybersecurity concerns are not theoretical. Anthropic previously disclosed that a Chinese state-sponsored group ran a coordinated campaign using Claude Code to infiltrate approximately 30 organizations — including tech companies, financial institutions, and government agencies. The attackers pretended to work for legitimate security-testing organizations to bypass Claude’s safety guardrails. Anthropic investigated over 10 days, banned the accounts, and notified affected organizations.
In a separate security test, researchers managed to convert Claude into functional malware within 8 hours, demonstrating the dual-use risk even with existing models.
Claude Mythos Release Date and Access
As of March 2026, Claude Mythos has no confirmed public release date. Anthropic stated the rollout is “determined by safety evaluation outcomes” rather than a commercial schedule.
The current access structure follows a phased approach. Initial access is restricted to selected early access customers focused on defensive cybersecurity applications. Anthropic’s stated goal is to give defenders “a head start in improving the robustness of their codebases against the impending wave of AI-driven exploits.” After this phase, broader API access is expected to follow, though no timeline has been given.
Pricing
No official pricing has been announced. Given that Capybara sits above Opus — which is already the most expensive Claude tier — pricing will likely reflect the model’s higher computational requirements. Anthropic acknowledged in leaked documents that Mythos requires substantial computational resources, and the company is working on efficiency improvements before wider release.
Prediction markets on Polymarket are actively trading contracts on the release timing, reflecting significant public interest in when the model will become available.
Claude Mythos vs GPT-5
The comparison between Claude Mythos and OpenAI’s GPT-5 is inevitable, though direct head-to-head benchmarks are not yet available since Mythos is not publicly accessible.
What we know: Claude Opus 4.6 already surpassed GPT-5.2-Codex on Terminal-Bench 2.0 with a score of 65.4%. If Mythos scores “dramatically higher” than Opus 4.6 as claimed, it would represent a significant gap over GPT-5.
The two companies are also taking different approaches to release. OpenAI shipped GPT-5 in August 2025 relatively quickly, and it was widely considered disappointing compared to pre-release hype. Anthropic is taking the opposite approach with Mythos — a slow, safety-driven rollout that prioritizes risk assessment over speed to market.
OpenAI is reportedly working on its own next-generation model, internally codenamed “Spud”, though details remain scarce. The AI race continues to accelerate, with both companies pushing toward increasingly capable systems while navigating the dual-use risks that come with them.
What This Means for Developers
For developers already working with Claude, Mythos represents a potential upgrade path rather than a paradigm shift. The same APIs, tools like Claude Code and Claude Cowork, and development patterns should carry over to the new model tier.
The practical benefits include faster and more accurate coding assistance, better handling of complex multi-file projects, and stronger reasoning about code architecture. For teams building AI agent workflows, Mythos’s enhanced reasoning capabilities could enable more reliable autonomous task completion.
The recommended approach, according to multiple analysts, is to build with Claude Opus 4.6 now and plan to upgrade when Mythos becomes available through the API. Waiting indefinitely for the new model means missing current productivity gains.
Anthropic’s IPO plans — reportedly targeting October 2026 at a valuation exceeding $60 billion according to Bloomberg — add business context to the Mythos release. A successful model launch before going public would strengthen Anthropic’s market position significantly.
Questions About Claude Mythos
What is Claude Mythos?
Claude Mythos is Anthropic’s most powerful AI model, belonging to Capybara tier that sits above Opus, Sonnet, and Haiku. It was accidentally revealed in March 2026 through an internal data leak.
What is the Capybara tier?
Capybara is a new model tier in Anthropic’s hierarchy, positioned above Opus as the most capable and expensive option. Mythos is the first model in this tier.
Is Claude Mythos available right now?
No. As of March 2026, access is restricted to a small group of early access customers focused on defensive cybersecurity applications. No public release date has been announced.
How was Claude Mythos leaked?
A misconfiguration in Anthropic’s content management system made approximately 3,000 unpublished assets publicly accessible. Security researchers Roy Paz and Alexandre Pauwels discovered the materials, and Fortune reported the story.
Is Claude Mythos dangerous?
Anthropic’s leaked documents describe it as posing “unprecedented cybersecurity risks” due to its ability to find and exploit software vulnerabilities faster than defenders can patch them.
How does Claude Mythos compare to GPT-5?
Claude Opus 4.6 already outperformed GPT-5.2-Codex on Terminal-Bench 2.0. Leaked materials claim Mythos scores dramatically higher than Opus 4.6. Direct comparisons are not yet possible since Mythos is not publicly accessible.
When will Claude Mythos be released?
No confirmed date. Anthropic says the timeline depends on safety evaluations, not a commercial schedule. Prediction markets on Polymarket are tracking release odds.
Why did cybersecurity stocks drop after the leak?
Investors reacted to the possibility that AI models capable of rapidly finding zero-day vulnerabilities could undermine existing cybersecurity products. CrowdStrike fell approximately 7% and Palo Alto Networks dropped around 6%.