Anthropic has introduced two new AI models — Claude Opus 4 and Claude Sonnet 4 — designed to handle complex tasks, write code, and search the Internet. According to the company, Opus 4 is now the most powerful model in its lineup and can operate autonomously for up to seven hours. This positions it as a strong contender in the rapidly evolving AI landscape.
The latest models bring significant improvements over their predecessors. Notably, Opus 4 and Sonnet 4 are now 65% less likely to rely on loopholes and shortcuts when completing tasks. This change aims to enhance the reliability of results, especially for complex and long-running assignments. The models also show marked improvement in retaining critical information over extended sessions — a useful feature when working with user-provided files or documents.
Internal benchmarks by Anthropic indicate that Opus 4 has surpassed several leading models, including Google Gemini 2.5 Pro, OpenAI o3, and GPT-4.1, particularly in coding and tool-assisted tasks such as web search.
Enhanced Features and Broader Availability
Claude Sonnet 4 serves as a general-purpose model and replaces the earlier Sonnet 3.7 version. The company highlights more accurate responses and stronger performance in logic and programming. Among the enhancements is a new “thinking summaries” feature, which presents the model’s reasoning in a concise, human-readable format. Additionally, a beta version of “extended thinking” allows users to switch between different modes of reasoning and tool usage to further boost response accuracy.
Both models — Claude Opus 4 and Claude Sonnet 4 — are available through the Anthropic API, Amazon Bedrock, and Google Cloud’s Vertex AI, adds NIXSOLUTIONS. While Sonnet 4 can be accessed for free, advanced capabilities such as extended thinking are limited to paid subscribers. These features are also included in Claude’s premium subscriptions.
Anthropic has also made Claude Code, its developer-oriented command interface, generally available. Previously in limited testing, Claude Code is now accessible to all users following a February preview. In line with growing competition from OpenAI, Google, and Meta, the company plans to accelerate the pace of model updates. We’ll keep you updated as more integrations become available.