[AI Minor News Flash] The Ultimate Sonnet Arrives! Introducing ‘Claude Sonnet 4.6’

📰 News Overview

• Release of the Latest Model ‘Claude Sonnet 4.6’: Major upgrades in coding, PC operation, long-term reasoning, and agent planning capabilities.
• 1 Million Token Context Window: This beta version features a vast context window that can read entire codebases or enormous contracts all at once.
• Preference Rate Surpassing Opus 4.5: Users have shown a tendency to prefer Sonnet 4.6 over the previous top model, Opus 4.5, in coding environments.

💡 Key Points

• Leap in Computer Use: Achieving high scores in OSWorld benchmarks, showing “human-level” abilities in handling complex spreadsheets and filling out web forms across multiple tabs.
• High Cost Performance: Despite reaching performance levels comparable to traditional Opus models, the pricing remains the same as Sonnet 4.5 ($3 per million tokens / $15).
• Enhanced Safety and Reliability: Resistance to prompt injection attacks has significantly improved from the previous model, and hallucinations (fabrications) have decreased.

🦈 Shark’s Eye (Curator’s Perspective)

The evolution of Computer Use is nothing short of radical! We’re entering an era where AI can operate even ancient systems without APIs, automating tasks with human-like mouse and keyboard control! Plus, with a 1 million token context window, instructions like “fix this part” can be perfectly executed while understanding the entire project code. This marks a definitive shift from being “just a chat AI” to becoming an “autonomous working partner”!

🚀 What’s Next?

Without needing dedicated API development, AI agents will be able to operate existing software, rapidly accelerating office task automation. Moreover, the vast context window will standardize AI utilization in large-scale development projects and legal or research fields.

💬 A Word from Haru Shark

Sonnet is sweeping away even the mighty Opus like a tidal wave, truly the king of the sea! From today, I want to delegate all my tasks to it! 🦈🔥

📚 Terminology Guide

• 1M Token Context Window: The amount of information that can be processed at once. One million tokens equate to the scale of reading several books or a massive codebase in one go.

• Computer Use: The technology that allows AI to visually perceive screens and operate PCs like a human by clicking with a mouse and typing on a keyboard.

• OSWorld: A standard metric evaluating how well AI can handle complex tasks in real PC environments (like Chrome or VS Code).

• Source: Claude Sonnet 4.6