Claude Sonnet 4.5: Anthropic's Leap in AI Coding and Safety Features
Anthropic's Claude Sonnet 4.5 launches with advanced coding features and improved safety measures.
- • Claude Sonnet 4.5 scores 61.4% on the OSWorld benchmark, surpassing its predecessor.
- • The model can code autonomously for up to 30 hours, enhancing productivity.
- • Pricing remains at $3 per million input tokens and $15 per million output tokens.
- • Improved safety features include classifiers for detecting harmful content.
Key details
Anthropic has made a significant update to its AI offerings with the launch of Claude Sonnet 4.5, touted as the company’s most advanced coding model to date. Released on September 30, 2025, this model stands out for its enhanced coding capabilities, allowing for the creation of production-ready applications. According to reports, Claude Sonnet 4.5 demonstrates impressive benchmarks, scoring 61.4% on the OSWorld benchmark, a substantial improvement over the 42.2% achieved by its predecessor.
The new model retains its pricing at $3 per million input tokens and $15 per million output tokens, consistent with earlier versions. It has been praised for its reliability and adaptability in various enterprise contexts, capable of autonomously coding for up to 30 hours without interruption. Mike Krieger, the Chief Product Officer at Anthropic, emphasized that Sonnet 4.5 targets both developers and office workers, facilitating a wider adoption of AI tools in professional settings.
Moreover, significant enhancements in safety have been integrated, including AI Safety Level 3 protections and classifiers designed to detect harmful content, aiming to reduce instances of problematic behaviors. This focus on AI safety is critical as the landscape of AI products becomes increasingly competitive, especially against rivals like OpenAI, which has recently made strides with its own coding models. The launch of Claude Sonnet 4.5 is a clear indication of Anthropic's strategic positioning in the rapid evolution of AI technology.