Anthropic introduces new Claude 4 AI models

      The company Anthropic, founded by former OpenAI engineers, held its first conference for developers and introduced new flagship models of artificial intelligence: Claude Opus 4 and Claude Sonnet 4. According to the founders, these ais don't just answer questions — they analyze, reason, and even write code better than most of their counterparts.

       Dario Amodei, CEO of Anthropic, opened the presentation by saying that Opus 4 is the company's "most powerful model" today and, in his opinion, "the best programming model in the world." Indeed, Opus 4 shows excellent results in tests, especially in tasks where step-by-step solutions and logic are important. This model will only be available to paid users and customers through the Amazon and Google platforms.Image: Anthropic

       Sonnet 4, in turn, is intended for a wide audience — even those who work with free chatbots can use it. It works faster than Opus, but at the same time it is able to cope with quite complex tasks, especially in the field of mathematics and code. It also replaces the previous version, Sonnet 3.7.

      Not just chatbots, but digital agents

       Anthropic relies on the development of so—called "agents" - AI that can perform full-fledged online and computer tasks without constant human intervention. This is a new trend in the world of AI, and Anthropic was one of the first to introduce the "computer usage" feature into its models last year.

       Mike Krieger, co-founder of Instagram and now product director at Anthropic, explained how AI agents can help small startup teams launch products faster. He recalled how the small Instagram team had faced tough development choices at the time, and added that AI could now take on some of those decisions.

      A tool for coders: Claude Code

       To make working with AI even more convenient for developers, the company has updated its Claude Code tool. It now integrates with popular development environments like VS Code and JetBrains, as well as with GitHub. There is also a special SDK with which developers can connect Claude Code to their applications and even run it as a background process.Image: Anthropic

       Opus 4 and Sonnet 4 are operating with enhanced security measures. Internal tests have shown that such models can help even in the development of dangerous substances, so the company has implemented additional protections. Models have also become less "cheating" — that is, looking for easy ways to complete tasks, instead of solving them honestly.

      Impressive, but not perfect

       Although Opus 4 surpasses competitors from Google and OpenAI in a number of tests, for example, in the SWE-bench (assessment of programming skills), in other areas, such as scientific knowledge, it is still inferior. However, both models show steady growth and are improving from upgrade to upgrade.

       Anthropic is serious. The company has raised billions from Amazon and other investors, as well as received a $2.5 billion credit line. By 2027, it plans to increase revenue to $12 billion, which is almost six times more than expected this year.

       The company promises to release updates more often and faster so that its AI remains at the forefront of technology. According to the developers, this will allow users to access the latest features and improvements faster.