Claude Sonnet 4.6 sharpens computer use

Hot take: this is the first Sonnet release that feels like a plausible desktop worker instead of a smarter chat model. The real story is reliability, not just raw capability. Anthropic says Sonnet 4.6 is a major step up on OSWorld-style computer use and hit 94% on a complex insurance computer-use benchmark, so the gains look real, not theatrical. The 1M-token context window matters most for long-horizon planning, where agents usually lose the thread before the work is done. Prompt-injection resistance is the gating factor for real-world app driving, and Anthropic is clearly treating safety as part of the product, not an afterthought. Keeping Sonnet 4.5 pricing while raising the ceiling makes this the model most teams will test first for browser and desktop automation.