Anthropic announced Claude Sonnet 5, a midsize foundation model that can make plans, operate browsers and terminals, and run autonomously—capabilities that previously required larger, costlier models. The company says the new model will be the default for both free and Pro subscriptions, giving developers immediate access without extra configuration.
Sonnet 5 delivers performance close to Anthropic’s flagship Opus 4.8 while undercutting its price. Through August 31 it costs $2 per million input tokens and $10 per million output tokens, rising to $3 per million input tokens afterward. On an agentic coding benchmark it scores 63.2%, up from Sonnet 4.6’s 58.1% and approaching Opus 4.8’s 69.2%; it also edges Opus 4.8 on a knowledge‑work test. “We handed Claude Sonnet 5 a two‑part job—update Salesforce tiers and send a launch announcement—and it finished end‑to‑end,” Zapier senior engineer Daniel Shepard said.
Safety‑wise, Sonnet 5 shows fewer undesirable behaviors, better refusal of malicious requests, and reduced hallucinations compared with its predecessor, though it still trails Opus 4.8. Co‑founder Fabian Hedin emphasized that a model that knows when to say no is as crucial as one that can build. The launch mirrors moves by OpenAI and Google, highlighting cost as the next competitive edge in agentic AI.
