Anthropic today officially released the new large language model Claude Sonnet 5, which is positioned as a more affordable mid-to-high-end model, significantly narrowing the gap in performance with the flagship Opus series. Officials stated that Sonnet 5 is the “most agent-capable” Sonnet model to date. It can independently plan tasks, call external tools such as browsers and terminals, and realize automated operations to a considerable extent.
Compared with the more expensive Opus model, the Sonnet series has always been priced at a wider user base, and this time Sonnet 5 is close to Opus 4.8 in terms of "agent capability" and overall performance. In key scenarios such as reasoning ability, tool use, code writing, and knowledge-based work, Sonnet 5 has significantly improved compared to the previous generation Sonnet 4.6.
In actual use, Sonnet 5 is emphasized to be able to complete some complex tasks that cannot be completed by Sonnet 4.6, and can self-check the output results without additional instructions. Anthropic also claims that the new model performs better at rejecting malicious requests, and the hallucination rate and sycophancy rates have also been reduced, thus further improving security and reliability.
In terms of product strategy, Sonnet 5 is available on all plans and becomes the default model for the Free plan and the Pro paid plan. Sonnet 5 is priced at $2 per million input tokens and $10 per million output tokens until August 31, after which the prices will increase to $3 and $15. Through this pricing strategy, Anthropic has reduced capabilities close to Opus 4.8 to a lower price range to cover the needs of a wider range of developers and ordinary users.