Anthropic, a leading artificial intelligence research company, has officially unveiled its latest AI model, Claude Sonnet 4.5, claiming it to be the “best coding model in the world.” Released on September 29, the model represents a major step forward in AI-driven coding, autonomous systems, and computer interaction, and is available to developers via API and to general users through the Claude website and apps.
Claude Sonnet 4.5 builds on the success of its predecessor, introducing significant upgrades in coding performance and autonomous capabilities. According to Anthropic, the model can operate continuously for over 30 hours without interruption — a remarkable improvement from the seven-hour limit of earlier versions. During these extended sessions, the model is capable of autonomously developing full applications, including configuring databases, purchasing domain names, conducting security audits, and more.
In benchmarks, Claude Sonnet 4.5 excelled, securing top performance on the SWE-bench Verified evaluation, a widely recognized test of real-world software engineering capabilities. This positions the model as a highly capable tool for developers, enabling end-to-end support in the software development lifecycle — from project planning and coding to debugging and deployment.
The model also made notable strides in computer interaction. It achieved a score of 61.4% on the OSWorld benchmark, which evaluates AI proficiency in handling real-world computer tasks — up from 42.2% in previous iterations. This performance boost allows Claude Sonnet 4.5 to efficiently manage complex tasks such as calendar scheduling, data analysis, document generation, and other automated workflows.
A significant addition to Claude Sonnet 4.5 is the Claude Agent SDK, designed to empower developers to build powerful autonomous agents. These agents can manage complex, long-duration tasks with minimal human supervision. The SDK opens up possibilities for applications in fields such as cybersecurity, finance, software engineering, and technical research, enabling the creation of systems that operate with unprecedented autonomy.
Anthropic has placed strong emphasis on safety and alignment in Claude Sonnet 4.5. The model underwent rigorous safety training aimed at reducing harmful behaviors, including sycophancy, deception, and power-seeking tendencies. Enhancements have also been made to safeguard against prompt injection attacks, ensuring safer deployment in real-world environments.
Claude Sonnet 4.5 is now available for developers through the Claude API at rates consistent with previous versions — $3 per million input tokens and $15 per million output tokens. The model is also accessible via Amazon Bedrock and Google Cloud’s Vertex AI, providing developers with flexible integration options for diverse platforms.
Industry experts suggest that Claude Sonnet 4.5 could redefine AI-assisted software development by combining advanced coding capabilities with autonomous agent functions, potentially transforming productivity in programming and other technical domains.
With its enhanced capabilities, Anthropic aims to position Claude Sonnet 4.5 as a leading choice for developers seeking intelligent, reliable, and autonomous AI tools for complex computing tasks.
