'World's Best Coding Model': Anthropic Unveils 'Claude Sonnet 4.5' To Enhance AI Reasoning

Anthropic has unveiled its latest AI model called Claude Sonnet 4.5 and claims it “is the best coding model in the world”. Supported by tech giants Alphabet and Amazon, Anthropic is in a race with other startups to build AI that can manage software and complete multiple steps of work reliably, bringing users closer to AI tools that can take on tasks in everyday life.

“Claude Sonnet 4.5 is state-of-the-art on the SWE-bench Verified evaluation, which measures real-world software coding abilities. Practically speaking, we’ve observed it maintaining focus for more than 30 hours on complex, multi-step tasks,” Anthropic said in a blog post.

According to the company, Claude 4.5 has improved capabilities in financial and scientific reasoning and demonstrates greater proficiency in computer use, achieving approximately 60% on a benchmark assessing operating system skills, compared to around 40% for earlier models.

“Experts in finance, law, medicine, and STEM found Sonnet 4.5 shows dramatically better domain-specific knowledge and reasoning compared to older models, including Opus 4.1,” the blog post said. 

Anthropic has rolled out a checkpoint feature for Claude Code, enabling users to easily return to previous code versions at any point in their work.

The Claude API has also been upgraded, allowing agents to operate for longer periods and manage more complex tasks. It is also now possible for users to create documents, presentations, and spreadsheets straight from a conversation with Claude.

Anthropic has unveiled ‘Imagine with Claude’, a tool that turns user descriptions into functioning software on the spot. By removing the need for preset syntax or coding skills, it lets people generate applications simply by writing what they want.

Microsoft has also announced it will incorporate new features into Microsoft 365 Copilot using Anthropic’s models. These include an "Agent Mode" for Excel and Word, as well as an "Office Agent" within Copilot chat, with similar enhancements planned for PowerPoint in the near future.