# Claude Mythos Preview

> **TL;DR.** Claude Mythos Preview is Anthropic's most capable frontier model, released April 7, 2026 as a gated research preview — not publicly available.

- **Category:** AI / Frontier Models / AI Safety
- **Stage:** validating
- **Age:** 70 days
- **Origin date:** 2026-04-07
- **First detected:** 2026-05-08
- **Canonical URL:** https://earlyterms.com/term/claude-mythos-preview
- **Sources:** 8 primary URLs

## Definition

Claude Mythos Preview is Anthropic's most capable frontier model, released April 7, 2026 as a gated research preview — not publicly available. It achieves 93.9% on SWE-bench Verified and demonstrates cybersecurity capabilities that surpass all but the most elite human security researchers.

Anthropic launched Mythos Preview alongside [Project Glasswing](https://www.anthropic.com/glasswing), an industry consortium with AWS, Apple, Microsoft, Google, CrowdStrike, JPMorgan Chase, and over 40 additional organizations. Access was deliberately restricted after the model autonomously found thousands of zero-day vulnerabilities — including a 27-year-old OpenBSD TCP bug and a 16-year-old FFmpeg flaw — in every major OS and browser.

## Analogy

Think of it as a master lockpick given only to locksmiths, not sold in hardware stores.

## Why it's emerging now

Anthropic announced Claude Mythos Preview on April 7, 2026 — their most capable model ever — but withheld general release due to cybersecurity risk. Project Glasswing, a coalition of 12 core tech giants plus 40+ partners backed by $100M in model credits, is using Mythos to defensively patch critical infrastructure before the capabilities proliferate to bad actors.

## Related terms

- *parent:* Claude Mythos
- *related:* claude-opus-4-6
- *related:* claude-opus-4-7
- *related:* Project Glasswing
- *related:* managed-agents
- *related:* claude-code
- *competitor:* gpt-5-5
- *related:* agentic-coding
- *related:* coding-agents
- *related:* ai-agent-traps
- *related:* context-window

## Sources

1. [Anthropic Red Team — Assessing Claude Mythos Preview's cybersecurity capabilities](https://red.anthropic.com/2026/mythos-preview/)
2. [Anthropic — Project Glasswing announcement](https://www.anthropic.com/glasswing)
3. [AWS Bedrock — Claude Mythos Preview model card (context window, model ID, launch date)](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-mythos-preview.html)
4. [Google Cloud Blog — Claude Mythos Preview on Vertex AI (private preview, Apr 8)](https://cloud.google.com/blog/products/ai-machine-learning/claude-mythos-preview-on-vertex-ai)
5. [UK AI Security Institute — Evaluation of Claude Mythos Preview's cyber capabilities](https://www.aisi.gov.uk/blog/our-evaluation-of-claude-mythos-previews-cyber-capabilities)
6. [Mozilla Hacks — Behind the scenes: Hardening Firefox with Claude Mythos Preview](https://hacks.mozilla.org/2026/05/behind-the-scenes-hardening-firefox/)
7. [Simon Willison — Anthropic's Project Glasswing (analysis)](https://simonwillison.net/2026/Apr/7/project-glasswing/)
8. [Hacker News — Project Glasswing (1,541 points, 836 comments)](https://news.ycombinator.com/item?id=47679121)

---
_Generated by EarlyTerms · https://earlyterms.com/term/claude-mythos-preview_