anthropic.com
|
ksl
|
|
Anthropic released Claude Sonnet 4.6, and the headline number is striking – early testers preferred it over Opus 4.5 in 59% of coding comparisons, mostly because it overengineers less and follows instructions more closely. Pricing stays at $3/$15 per million tokens, same as Sonnet 4.5, with a 1M token context window now in beta. Computer use reaches what Anthropic calls human-level on tasks like multi-step form filling and spreadsheet navigation, though it still trails expert users on OSWorld benchmarks. Prompt injection resistance improved substantially over Sonnet 4.5. The release fits a pattern visible across OpenAI and Google as well: mid-tier models closing the gap on flagship offerings fast enough that the pricing hierarchy starts to look awkward.
