Anthropic presented this Tuesday (25) the Claude Sonnet 5, the latest evolution of its most sought-after artificial intelligence model. According to the company, this new version stands out as the most competent in its class for the development of agentic AI and the construction of autonomous systems.
Sonnet 5’s design was designed to enable AI to formulate strategies, handle resources such as internet browsers and command line interfaces, and complete activities completely independently. This capability represents a notable leap forward compared to previous iterations, which often encountered difficulties handling more intricate missions.
The model demonstrates impressive performance, being able to self-evaluate its own results without relying on direct instructions. This ability to self-reflect is a crucial differentiator for agentic AI, as it reduces errors and optimizes efficiency, placing it at a performance level very close to that of Opus 4.8, Anthropic’s most sophisticated algorithm.
Surprisingly, in certain knowledge assessments, Sonnet 5 even outperformed Opus 4.8. It also offers notable improvements compared to Sonnet 4.6, especially in logical reasoning, coding and interaction with different tools, providing even greater security for autonomous applications.
However, its ability to carry out high-risk tasks is inferior when compared to other models from the company, such as the Mythos 5. Despite this, the Sonnet 5 is equipped with protection mechanisms that are activated by default, with the purpose of identifying and preventing potentially dangerous actions in real time.
Integrated as standard in Claude’s free and Pro packages, Sonnet 5 has an initial cost of $2 per million tokens processed on input and $10 per million tokens generated on output. These promotional values are only valid until August 31st, after which they will be adjusted to US$3 and US$15 per million tokens, respectively.
