About this video
Anthropic has just confirmed Claude Mythos — a frontier AI model so capable they're refusing to release it to the public. We only found out because of an embarrassing data leak in March that exposed draft blog posts referencing the model. Two weeks later, Anthropic launched Project Glasswing — a coalition with Apple, Google, Microsoft, AWS, Cisco, JPMorgan and others to use Mythos for defensive cybersecurity. In this video I break down what Mythos actually is, why the benchmarks are a step change above Opus 4.6 (93.9% on SWE-bench Verified is genuinely shocking), the 27-year-old OpenBSD bug it found autonomously, and what this means for every developer over the next 12 months. 📌 KEY TAKEAWAYS: - 🦋 Project Glasswing explained — and why 12 of the world's biggest tech companies signed on - 📊 The benchmark gap between Mythos and Opus 4.6 (it's not subtle) - 🐛 The 27-year-old OpenBSD bug and 16-year-old FFmpeg flaw Mythos found on its own - 🔐 Why Anthropic is deliberately holding this model back from public release - 💰 The $25/$125 per million token pricing and what it tells us about who this is for - ⚠️ The dual-use problem and what it means for your codebase in 2026 - 🚀 What to expect from the next Opus model and why it matters