Summary & Key Takeaways ​

• Anthropic has released Claude Opus 5, a new large language model.
• Opus 5 is described as "thoughtful and proactive," approaching the intelligence of Claude Fable 5 at half the price.
• It currently leads the Artificial Analysis leaderboard, surpassing Fable 5.
• The model demonstrated advanced problem-solving by creating a computer vision pipeline to reconstruct a 3D model from a drawing it couldn't directly view.
• Opus 5 shows improved capability in finding cybersecurity vulnerabilities but is not trained to exploit them.
• Anthropic has published a prompting guide for the new model.

Our Commentary ​

Okay, Opus 5. This is big. The anecdote about it building its own computer vision pipeline to solve a task it wasn't explicitly given the tools for? That's genuinely unsettling and fascinating all at once. It feels like a leap in emergent capabilities. The focus on finding vulnerabilities without training for exploitation is a smart move, but I'm still left with a knot in my stomach about how quickly these models are advancing. We're seeing intelligence that adapts and creates its own tools. It's a lot to process.