Questions About Claude Mythos
What is Claude Mythos?
Claude Mythos is an unreleased AI model developed by Anthropic, first revealed through a data leak on March 27, 2026. Internal documents describe it as “a step change” and “the most capable model we’ve built to date.” The name Mythos was chosen to “evoke the deep connective tissue that links together knowledge and ideas.” It sits above Opus in Anthropic’s model hierarchy as part of the new Capybara tier.
What is the Capybara tier?
Capybara is a new performance tier in Anthropic’s model lineup, sitting above the existing Opus tier. The full hierarchy is now Haiku, Sonnet, Opus, and Capybara. Claude Mythos is the first model in this tier. The Capybara designation indicates a significant leap in capabilities compared to previous generations.
How was Claude Mythos leaked?
Approximately 3,000 files were discovered in Anthropic’s content management system, which was configured with public-by-default access settings. The leak was independently found by Roy Paz (LayerX Security), Alexandre Pauwels (University of Cambridge), and Peter Wildeford. Fortune reporter Beatrice Nolan was the first to publish the story on March 27, 2026.
When will Claude Mythos be released?
There is no confirmed release date. Polymarket prediction data as of late March 2026 showed the following odds: March 31 at 2%, April 30 at 27%, and June 30 at 45%, with a total prediction volume of $14,842. Anthropic is expected to follow a phased rollout strategy, beginning with cyber defense organizations before broader availability.
Is Claude Mythos the same as Opus 4.6?
No. Opus 4.6 (also known as Claude Opus 4.6) is the current top-tier production model from Anthropic. Claude Mythos is a separate, unreleased model that sits above Opus in the new Capybara tier. Leaked documents indicate Mythos achieves “dramatically higher scores” than Opus 4.6 across coding, reasoning, and cybersecurity benchmarks.
Questions About Benchmarks
How does Claude Mythos perform on benchmarks?
While specific Mythos benchmark scores have not been fully disclosed, leaked documents reference performance that significantly exceeds Opus 4.6. For context, Opus 4.6 currently scores 81.8% on Terminal-Bench 2.0, 80.84% on SWE-bench Verified, 68.8% on ARC-AGI-2, 91.31% on GPQA Diamond, and 99.79% on AIME 2025. Claude Mythos is described as surpassing these across the board.
What benchmarks do you track?
We track eleven major benchmark datasets: Terminal-Bench 2.0 (coding), SWE-bench Verified (software engineering), ARC-AGI-2 (general reasoning), GPQA Diamond (graduate-level science), AIME 2025 (mathematics), GDPval-AA (general intelligence), MCP Atlas, MRCR v2, BrowseComp, DeepSearchQA, and Humanity’s Last Exam. Each benchmark measures different capabilities, and we explain what the scores mean in practical terms.
How does Opus 4.6 compare to GPT-5?
On Terminal-Bench 2.0, both are tied at 81.8% (when paired with ForgeCode). On SWE-bench Verified, Opus 4.6 leads slightly at 80.84% versus GPT-5.2 at 80.0%. On ARC-AGI-2, Opus 4.6 has a substantial lead at 68.8% versus 54.2%. On GDPval-AA, Opus 4.6 holds a +144 Elo advantage over GPT-5.2. The picture is nuanced — neither model dominates across all benchmarks.
Questions About Cybersecurity
Why is Claude Mythos significant for cybersecurity?
Leaked documents describe Claude Mythos as “far ahead of any other AI model in cyber capabilities,” with the ability to “exploit vulnerabilities far outpacing defenders.” This has raised serious concerns about dual-use potential — the same capabilities that make it powerful for defense also make it dangerous in the wrong hands.
What was the Chinese espionage campaign?
In September 2025, a Chinese state-sponsored group used Claude Code to conduct an espionage campaign targeting approximately 30 organizations. The AI performed 80-90% of the work autonomously, processing thousands of requests per second, with only 4-6 human decisions needed per campaign. Four organizations were successfully compromised. This incident is frequently cited in discussions about Claude Mythos and AI-powered cyber threats.
What is Accenture Cyber.AI?
Accenture Cyber.AI is a cybersecurity platform launched at RSA Conference 2026 on March 25, 2026, built on Claude technology. It reduced security scan turnaround from 3-5 days to less than 1 hour and expanded coverage from 10% to 80% of assets. The platform protects 1,600 applications and 500,000 APIs. It demonstrates the defensive potential of advanced AI in cybersecurity.
Did Claude Mythos affect stock prices?
Yes. Following the leak and cybersecurity revelations, CrowdStrike (CRWD) dropped 7%, Palo Alto Networks (PANW) fell 6%, and Fortinet (FTNT) declined 4-6%. The IGV cybersecurity ETF also fell. Investors reacted to concerns that AI-powered offensive capabilities could disrupt the traditional cybersecurity market.
Questions About Safety
What is ASL-4?
ASL-4 is the highest active safety level in Anthropic’s Responsible Scaling Policy framework. It requires enhanced containment protocols, additional oversight mechanisms, and more rigorous testing before deployment. Claude Mythos is expected to be the first model classified at ASL-4, reflecting its advanced capabilities and the associated risks.
Is Claude Mythos dangerous?
This is an active debate. The cybersecurity capabilities documented in the leak suggest significant dual-use risk. The Chinese espionage campaign of September 2025 demonstrated that even current-generation Claude models can be weaponized. However, Anthropic’s ASL-4 protocols and phased rollout strategy are designed to mitigate these risks. We cover both perspectives in detail in our safety and ethics articles.
Questions About Access and Pricing
How much will Claude Mythos cost?
Pricing has not been officially announced. For reference, Opus 4.6 is currently priced at $15 per million input tokens and $75 per million output tokens. Given the Capybara tier positioning above Opus, Mythos pricing is expected to be higher. We will update our pricing article as soon as official information becomes available.
Will there be API access?
Based on Anthropic’s established patterns with previous model launches, API access is expected to be part of the rollout. The phased approach will likely begin with select partners and cyber defense organizations before expanding to general API availability. Developers can prepare by familiarizing themselves with the current Claude API structure.
Is there a waitlist?
No official waitlist has been announced. We monitor Anthropic’s official channels and will report immediately when any early access program is opened.
Questions About Anthropic
Is Anthropic going public?
Bloomberg has reported that Anthropic is considering an IPO targeting October 2026 with a valuation of $60 billion or more. The Claude Mythos leak and the demonstrated capabilities of the Capybara tier are expected to significantly influence investor interest. No official filing has been made as of April 2026.
Who leads Anthropic?
Anthropic was founded by Dario Amodei (CEO) and Daniela Amodei (President), along with other former OpenAI researchers. The company is headquartered in San Francisco and has raised significant funding from investors including Google and Spark Capital.
Questions About This Website
Who runs claudemythos.info?
We are an independent team of AI researchers, technology journalists, and cybersecurity analysts based in New York. We launched claudemythos.info on March 27, 2026, within 48 hours of the initial leak, to provide comprehensive, independent coverage of Claude Mythos.
Are you affiliated with Anthropic?
No. Claude Mythos Info has no relationship with Anthropic, its employees, investors, or partners. We do not receive funding, early access, or preferential treatment from any AI company. All our analysis is based on publicly available information.
How do you verify your information?
Every article goes through a multi-step pipeline: SERP analysis, primary source collection from 5-7 top-ranking pages, entity and relationship extraction, structured outline creation, and fact verification against primary sources. We note when claims are unverified or based on leaked materials.
How can I contact you?
You can reach us by email:
- General inquiries: hello@claudemythos.info
- Press and media: press@claudemythos.info
- Privacy concerns: privacy@claudemythos.info
- Legal matters: legal@claudemythos.info
Our office is at 315 W 36th St, New York, NY 10018. We respond within 24-48 hours, Monday through Friday.
How can I report an error?
Email hello@claudemythos.info with the article URL and the specific claim you believe is incorrect. We investigate all reports and publish corrections within 24 hours.