Claude 4.5 Sonnet is a multimodal language model developed by Anthropic, designed to balance high-level intelligence with operational efficiency. Released as a significant update to the Claude 4 series, it is optimized for complex software engineering, autonomous agent behavior, and large-scale analytical tasks. The model supports a 200,000-token context window and a 64,000-token output limit, enabling it to process and generate extensive technical documentation and codebases.
A primary feature of the model is its hybrid reasoning architecture, which includes an "Extended Thinking" mode. This capability allows the model to toggle between standard low-latency responses and a deeper chain-of-thought process for difficult problems. When active, this mode provides increased transparency by exposing the model's internal reasoning steps, which helps users verify the logic behind its conclusions in domains like mathematics and scientific research.
In addition to text-based reasoning, Claude 4.5 Sonnet features advanced computer use capabilities. It is designed to navigate graphical user interfaces, interact with web browsers, and execute multi-step workflows across different applications. Anthropic reports substantial improvements in agentic reliability, with the model demonstrating the ability to maintain focus on complex, autonomous tasks for extended durations exceeding 30 hours.
Safety and alignment are integrated through upgraded classifiers and adherence to AI Safety Level 3 (ASL3) protocols. The model exhibits reduced tendencies toward sycophancy and deception compared to previous versions, while maintaining high epistemic robustness when handling ambiguous or contextually shifted prompts. It is provided with enhanced defenses against prompt injection and other adversarial attacks to support secure enterprise automation.