Software Development

Anthropic launches Claude Sonnet 4.5 for production-ready coding

Ken Metral

29 Sep 2025 — 2 min read

Credit: Anthropic PBC

Anthropic has unveiled Claude Sonnet 4.5, its latest frontier AI model, which it claims delivers industry-leading performance on software engineering tasks and can autonomously build “production-ready” applications.

A Leap Beyond Prototyping

Announced Monday, Claude Sonnet 4.5 is available immediately via the Claude API and Claude chatbot, at the same pricing as its predecessor: $3 per million input tokens and $15 per million output tokens.

Anthropic says the new model represents a significant leap from previous generations, moving beyond quick prototyping to building full, reliable applications. In enterprise trials, Claude Sonnet 4.5 reportedly coded autonomously for up to 30 hours, setting up database services, purchasing domain names, and even conducting a SOC 2 audit to ensure security.

Benchmarks and Developer Praise

The company says Claude Sonnet 4.5 outperforms competitors on several major coding benchmarks, including SWE-Bench Verified. Still, Anthropic researcher David Hershey noted that benchmarks don’t fully capture the model’s abilities, especially on long-horizon tasks.

Early partners in the developer community are already impressed:

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

Competing With OpenAI

Over the past year, Anthropic’s Claude models have become popular among developers and enterprises, especially for software engineering tasks. Tech giants like Apple and Meta reportedly use them internally, and companies such as Cursor, Windsurf, and Replit integrate Claude via API.

However, OpenAI’s GPT-5 has recently outperformed Claude on some coding benchmarks, raising the competitive stakes. Anthropic is betting Claude Sonnet 4.5 will re-establish its leadership in coding-focused AI.

Safety and Alignment Improvements

Anthropic also emphasized that Claude Sonnet 4.5 is its most aligned model to date, with lower rates of sycophancy (over-agreeableness) and deception. The company says it has also strengthened defenses against prompt injection attacks, a growing concern in enterprise AI deployments.

New Tools for Developers

Alongside the model release, Anthropic introduced:

Claude Agent SDK – the same infrastructure powering Claude Code, now available for developers to build their own AI agents.
Imagine with Claude – a temporary research preview for Max subscribers, showing real-time code generation without prewritten functionality.

The Bigger Picture

Claude Sonnet 4.5 underscores Anthropic’s ambition to stay at the forefront of AI coding models. By promising production-ready reliability, enterprise-grade safety, and new developer tools, the company is positioning itself not just as a model provider, but as a platform for building the next generation of AI-driven applications.

Amazon debuts redesigned Kindle Scribe, including a color version with AI

At its annual fall hardware event on Tuesday, Amazon introduced two major updates to its Kindle lineup: a redesigned Kindle Scribe and the company’s first-ever color e-reader, the Kindle Scribe Colorsoft. Both devices mark a push toward blending the familiar paper-like Kindle experience with new AI-powered features. A Bigger,

DoorDash launches Dot, a small autonomous delivery robot rolling out in Phoenix

On Tuesday, DoorDash pulled back the curtain on Dot, a small, bright-red delivery robot designed to ferry food and small packages across city streets, bike lanes, and sidewalks at speeds of up to 20 miles per hour. Built entirely in-house, Dot looks more like a Pixar sidekick than a futuristic

Wealthfront files for U.S. IPO with $88 billion in assets and 1.3 million customers

Wealthfront, the fintech startup known for pioneering the robo-advisor approach to automated investing, has officially filed for a U.S. initial public offering (IPO). The move positions Wealthfront among a growing wave of fintech firms heading to the public markets this year, following the likes of Chime and Klarna. From

DeepSeek releases 'Sparse Attention' model that cuts API costs by 50%

Researchers at DeepSeek have introduced a new experimental AI model, V3.2-exp, designed to dramatically reduce inference costs during long-context operations. The announcement was made Monday via a post on Hugging Face, accompanied by a linked academic paper published on GitHub. A New Take on Attention The standout feature of