Sonnet 4.6 is Anthropic’s latest release, and it’s basically the story of “Opus-level performance at Sonnet pricing”—especially for coding. In this video I walk through what’s new, why Anthropic is clearly targeting knowledge workers + developers, and why I think this model continues their pattern of shipping state-of-the-art coding capability. The headline upgrades are computer use improvements, stronger long-context reasoning with a 1M context window, and better agentic planning. I also cover the web search update (dynamic filtering), which matters if you’re building agents that need to retrieve and validate information in a browser. I get the benchmarks out of the way (it’s very close to Opus 4.6 on key evals, with some lag in certain agentic search/browser scenarios), then I talk pricing—because it’s not as simple as “cheap Sonnet.” Under 200k tokens the pricing is familiar, but above 200k tokens the cost jumps hard (think 5x), so you need to architect around that if you’re doing long-context workflows. Finally, I run practical tests: a web dev build, a 500-star gravity simulation with click-to-add black holes, and a real “computer use” style UI testing flow inside Claude Code for a mission-control style agent system. The takeaway: Sonnet 4.6 is a beast for coding and UI-driven automation, but you still need to prompt and iterate for design quality—and you need to respect the context pricing cliffs.