Yesterday marked the grand unveiling of Claude Sonnet 4.5, a model that's quickly earning accolades as the premier coding solution globally.

This advanced system excels at constructing intricate Agents and showcases marked enhancements in reasoning and mathematical prowess when compared to its predecessor, Sonnet 4.

Claude Sonnet 4.5 has soared to the pinnacle of the SWE-bench Verified assessment, a benchmark designed to gauge software coding proficiency. Notably, it has demonstrated the ability to sustain concentration for over 30 hours while tackling complex, multi-faceted tasks.

In a rigorous benchmark test evaluating AI models' prowess in real-world computer tasks, Sonnet 4.5 emerged as the leader, scoring an impressive 61.4%. In contrast, Sonnet 4 achieved a score of 42.2%.

When subjected to reasoning and mathematics evaluations, Sonnet 4.5 outperformed competitors such as GPT-5 and Gemini-2.5, boasting a flawless 100% accuracy rate.

Within specialized domains like finance, law, medicine, and STEM fields, Sonnet 4.5 exhibited unparalleled domain-specific expertise and reasoning abilities, securing a 72% victory over Opus-4.1.

Cursor and GitHub users have showered it with glowing praise:

Sonnet 4.5 also comes packed with an array of innovative features inherited from the Claude Code coding Agent. These include seamless access to virtual machines and memory, enhanced context management, and robust multi-Agent support.

Anthropic has announced that Sonnet 4.5 is the inaugural model in their lineup capable of reconstructing the Claude.ai web application from scratch. This monumental task, which spanned approximately five and a half hours, involved the utilization of over 3,000 tools.

Priced identically to Sonnet 4, users can now indulge in a superior coding experience without any additional cost.

Presently, Anthropic's coding Agent leverages this cutting-edge model. Claude Code has already amassed over $500 million in operational revenue, with its user base surging more than tenfold in the past three months. A native Visual Studio Code extension is also on the horizon. Developers can now monitor changes implemented by Claude Code in real-time through inline diffs.

The terminal version of Claude Code has also undergone enhancements, including improved status visibility and a searchable prompt history.

When Claude Code encounters anomalies, there's no longer a need to manually integrate code into the codebase or perform local backups. Simply undo the changes with ease.

For developers embarking on Agent construction, Anthropic has introduced the Claude Agent SDK. This new SDK, built on the same robust infrastructure as Claude Code, empowers developers to craft any Agent they envision. It encompasses features such as Agent orchestration, memory and context management, tool utilization, and permission control.

On the API front, Anthropic has integrated an automatic context management feature, enabling Claude to intelligently edit the context window and purge outdated data as necessary.

The team has also conducted comprehensive security training for Sonnet 4.5, refining the model's behavior and curbing undesirable tendencies such as flattery, deception, and power-seeking.

In recent months, numerous AI experts have engaged in discussions about harnessing artificial intelligence to create the desired software. Sonnet 4.5 exemplifies the potential of building Agents and charts a promising course for the future.

References: https://www.anthropic.com/news/claude-sonnet-4-5