The company Anthropic introduced a new AI model Claude Sonnet 4.5, which is now available through the Claude API and in the Claude chatbot. Developers can use the same pricing as the previous version: $3 per million input tokens and $15 per million output tokens. The model is notable for its ability to operate autonomously for up to 30 hours and create full-fledged applications, including database setup, domain name purchases, and security audits.
Claude Sonnet 4.5 demonstrates leading results in programming tests such as SWE-Bench Verified and also performs better in long-term planning and complex tasks in the fields of cybersecurity, financial services, and research. Representatives from Canva noted that the model helps solve complex development tasks, integrate new features, and support research in their products.
Along with the launch of Claude Sonnet 4.5, Anthropic introduced the Claude Agent SDK, which allows developers to create their own agents based on the same infrastructure used in Claude Code. The company also opened temporary access to the “Imagine with Claude” feature for Max subscribers, where the model generates software in real-time in response to user requests.
Anthropic developers report that Claude Sonnet 4.5 has become more resistant to prompt attacks and less prone to repetition or false responses. The model has significantly improved computer usage skills: it can now work with browsers, analyze data, schedule meetings on calendars, and create tables with information found in open sources.
