Claude Mythos: AI Completes 16-Hour Tasks 50% of Time

Claude Mythos, a new AI model, can now complete complex software tasks that would take a human expert over 16 hours, succeeding half the time. This is a huge leap forward for AI capabilities. Here's the thing: The AI safety organization METR tested Mythos and found it pushed past the limits of their current measurement tools. Earlier models like GPT-4o could only manage tasks of about 7 minutes, while Claude Opus 4.6 reached around 5 to 6 hours. Mythos is significantly more advanced. What's interesting is that this "time horizon" measures task difficulty, not how long the AI takes. It shows that Mythos can reliably handle very challenging, self-contained technical work. This rapid progress, doubling every 105 days, highlights how quickly AI is evolving. The bottom line: This shows AI is becoming incredibly adept at complex technical tasks, potentially transforming industries like software engineering and cybersecurity.