Anthropic launches Claude Sonnet 4.5, its best AI model for coding

Anthropic has unveiled Claude Sonnet 4.5, its latest frontier AI model, which the company claims delivers state-of-the-art performance on coding benchmarks and can build “production-ready” applications rather than just prototypes — a significant leap in reliability and capability.

The new model is now available through both the Claude API and Claude chatbot, at the exact pricing as Claude Sonnet 4: $3 per million input tokens (roughly equivalent to 750,000 words, or more than the length of The Lord of the Rings) and $15 per million output tokens.

A Major Step Forward for AI Coding Models

Over the past year, Anthropic’s Claude models have become favourites among developers and enterprises thanks to their strong performance in software engineering and automation tasks. Tech giants like Apple and Meta reported utilising Claude internally, while Anthropic has expanded its business by providing API access to coding tools such as Cursor, Windsurf, and Replit.

However, the company’s dominance has been increasingly challenged by OpenAI’s GPT-5, which recently outperformed Claude models on several coding benchmarks.

Anthropic says Claude Sonnet 4.5 now leads the industry again, particularly in benchmarks such as SWE-Bench Verified, which measures coding reliability and execution accuracy.

Image Credits: Anthropic

Still, David Hershey, an Anthropic AI researcher, told TechCrunch that benchmarks alone don’t fully capture the model’s real-world performance:

“I’ve seen Claude Sonnet 4.5 code autonomously for up to 30 hours during early enterprise trials,” Hershey said. “In that time, it not only built an application but also stood up database services, purchased domain names, and even performed a SOC 2 audit to ensure the product was secure.”

Industry Praise for Claude Sonnet 4.5

Cursor CEO Michael Truell called Claude Sonnet 4.5 “state-of-the-art” for long-horizon coding tasks, while Windsurf CEO Jeff Wang described it as “a new generation of coding models.”

Anthropic says Sonnet 4.5 also brings more substantial model alignment, featuring lower levels of sycophancy and deception than its predecessors, and better resistance to prompt injection attacks — a growing concern for enterprise users.

New Tools and Research Features

Alongside the model launch, Anthropic introduced the Claude Agent SDK, a new developer toolkit that leverages the same infrastructure powering Claude Code. The SDK enables developers to build their own autonomous agents, further expanding Anthropic’s ecosystem for enterprise AI development.

The company also rolled out a temporary research preview called “Imagine with Claude” for Claude Max subscribers. The experimental feature enables users to watch the AI generate software in real-time, responding dynamically to user prompts without requiring prewritten code or fixed templates.

A Rapidly Evolving AI Race

The release of Claude Sonnet 4.5underscoress the accelerating pace of competition in the AI industry, where leading firms are rapidly introducing their flagship models. The new model arrives less than two months after Anthropic’s previous release, Claude Opus 4.1, reflecting the company’s drive to maintain momentum in the face of fierce competition from OpenAI, Google DeepMind, and others.

With Claude Sonnet 4.5, Anthropic aims to set a new benchmark for AI coding reliability, combining long-context reasoning, tool integration, and a new level of autonomy in software creation.