Anthropic представляет Claude Sonnet 4.5 — ИИ, который меняет правила игры в программировании Translation: Headline: Anthropic unveils Claude Sonnet 4.5 — the AI that is changing the game in programming

Anthropic has introduced its latest AI model, Claude Sonnet 4.5, showcasing cutting-edge capabilities in the programming domain. According to the company, this model is proficient in generating «production-ready» applications.

«This is the most consistent frontier model we have ever released, featuring significant enhancements across various areas compared to earlier versions of Claude,» the blog states.

In the past year, solutions from Anthropic have become popular choices for tackling programming challenges. Reports indicate that companies like Apple and Meta are utilizing Claude models within their organizations.

The startup claims that Claude Sonnet 4.5 demonstrates industry-leading performance across several programming benchmarks, including the SWE-Bench Verified.

The model can autonomously generate code for up to 30 consecutive hours. During one test, it not only developed an application but also deployed database services, registered domain names, and conducted an audit based on the SOC 2 standard.

With the launch of Claude Sonnet 4.5, Anthropic also unveiled the Claude Agent SDK, a framework for creating agents.

Other updates include the demonstration of how the Claude extension operates directly within browsers, enabling it to navigate websites, fill out spreadsheets, and perform tasks.

Anthropic noted significant advancements in protection against prompt injection attacks, where an attacker embeds hidden instructions into prompts to manipulate the model’s behavior in unsafe or rule-violating ways.

The neural network will be accessible through an API and a Claude chatbot, with pricing set at $3 per million input tokens and $15 per million output tokens.

Recall that in August, Anthropic programmed the Claude Opus 4 and 4.1 chatbots to conclude conversations with users «in rare, extreme cases of systematic harmful or abusive interactions.»