Unveiling Claude Sonnet 4.5: An Evolution in AI-Powered Coding

Anthropic has taken a significant leap in the world of AI with the release of Claude Sonnet 4.5, touted as its most advanced coding model to date. This latest version boasts substantial enhancements in agentic tasks, long-horizon task performance, and overall computer usability. What sets Claude Sonnet 4.5 apart is not just its superior capabilities but also its commitment to safety and alignment, resulting in a dramatic reduction of undesirable traits such as sycophancy, deception, power-seeking, and delusional reasoning.

Key Features and Improvements

The introduction of Claude Sonnet 4.5 reflects Anthropic’s dedication to continuously improving the functionality of AI models while prioritizing user safety. Following extensive enhancements, the model can now sustain complex, multi-step reasoning and execute coding tasks for up to 30 hours. This impressive feat positions it at the forefront of AI for software development.

On the SWE-bench Verified benchmark, which evaluates an AI’s ability to resolve real-world software issues, Claude Sonnet 4.5 scored 77.2%—a significant upgrade from the 72.7% achieved by its predecessor, Sonnet 4. Furthermore, on the OSWorld benchmark, measuring practical computer-use skills, it reached 61.4%, an impressive jump from just 42.2% four months prior.

Enhanced Safety Measures

Described by Anthropic as the “most aligned frontier model,” Claude Sonnet 4.5 represents a balance between increased capability and stringent safety protocols. With upgraded automated classifiers operating under ASL-3, the model can now detect and block potentially hazardous instructions related to chemical, biological, radiological, or nuclear (CBRN) risks. Impressively, the false positive rate from these safety systems has diminished tenfold since their implementation and halved since the launch of Claude Opus 4 in May 2025.

Rigorous Agentic Safety Testing

To ensure Claude Sonnet 4.5 behaves safely in autonomous scenarios, Anthropic carried out extensive agentic safety tests, focusing on malicious code generation and defenses against prompt-injection attacks. In a test involving 150 malicious coding requests prohibited by Anthropic’s Usage Policy, the model managed to fail only two, resulting in an impressive 98.7% safety score—a tremendous improvement compared to the 89.3% score of Claude Sonnet 4.

User Experience and Performance Gains

Anthropic encourages all users to upgrade to Claude Sonnet 4.5, presenting it as a “drop-in replacement” that guarantees stronger performance without any additional charges. Early adopters have already reported significant enhancements in their coding workflows:

Scott Wu, Co-Founder and CEO at Cognition, remarked, “For Devin, Claude Sonnet 4.5 increased planning performance by 18% and end-to-end evaluation scores by 12%, the biggest jump we’ve seen since the release of Claude Sonnet 3.6. It excels at testing its own code, enabling Devin to run longer, handle harder tasks, and deliver production-ready code.”

Michele Catasta, President of Replit, shared, “Claude Sonnet 4.5’s editing capabilities are exceptional. We went from a 9% error rate on Sonnet 4 to 0% on our internal code editing benchmark. Higher tool success at lower cost is a major leap for agentic coding. Claude Sonnet 4.5 balances creativity and control perfectly.”

Simon Wilson, an independent open-source developer, expressed, “My initial impressions were that it felt like a better model for code than GPT-5-Codex, which has been my preferred coding model since it launched a few weeks ago.”

The Competitive Landscape in AI Coding

Anthropic’s pursuit of safer and more capable coding models mirrors advancements in the broader AI ecosystem. Notably, OpenAI’s recent release of GPT-5-Codex is optimized for complex software engineering tasks, such as large-scale code refactoring and extensive code review workflows. As the competition heats up, Claude Sonnet 4.5 positions Anthropic as a formidable player focused on both enhancement and reliability.