Anthropic has launched two new AI models: Cloud Opus 4 and Cloud Sonnet 4. Anthropic says these models are the best in the industry, and their highlight is their ability to argue. In addition, both Opus 4 and Sonnet 4 are designed to improve tasks such as coding and agent. According to Anthropic, Cloud Opus 4 is the most powerful model to date and aims to work on long and complex tasks for developers. Anthropic writes on his blog post, “Claude Opus 4 is the world’s best coding model, with complex, long -running tasks and continuous performance on agent workflows.”

Meanwhile, Cloud Sonnet 4 is described as its predecessor, Cloud Sonnet 3.7 as practical, efficient upgrade, and now it is available to free-level users. Opus 4 will be available only to the customers paid for clouds.

According to Anthropic, one of the major highlights, the coding is a strong performance of Ops 4 on the benchmark. The AI ​​company claims that it scored 72.5 percent on SWE-BEND and 43.2 percent on Terminal-Bench. Which basically means that the model can work for several hours at a time without the alleged release of performance, making it suitable for projects that require constant attention.

Cloud Sonnet 4 also shows improvement in previous versions, scoring 72.7 percent on SWE-Bench. Although it does not match Opus 4 in overall capacity, Anthropic states that it creates a better balance between speed and accuracy, which makes it suitable for broad, more everyday tasks.

Both models come with “extended thinking” and anthropic calls using equipment. This means that they can prevent logic, use equipment such as web search or code execution, and then resume their thought process. The tool can now be used in parallel, which helps with more complex workflows.

Models also introduce new memory features. If access to local files, they can remove the major facts and save them for future use. Anthropic states that it helps the model to make better long -term memory and improves performance on functions that require continuity.

Anthropic has also announced four new API facilities, which are also rolling out. This includes a code execution tool, a connector for multi-component programs (MCP) of anthropic, a file API and Prompt Caching for up to one hour. These updates make it easy for developers to manufacture AI agents that can take more complex tasks.

Anthropic claims that Cloud Oppus 4 shows strong memory performance, especially in agents -like settings. Anthropic says that it gave AI model file access during a game of Pokémon, and was able to create “navigation guide” and maintain awareness about previous tasks. It also reduces the possibility of using shortcuts or flaws to complete tasks, some more prone to previous models.

Anthropic has added a new feature called “Thinking Summary” to help users understand how models reach conclusions. These are small interviews of the model’s logic process, produced by a small AI model. The request at the request through a developer mode is still available.