Meet Claude 4 Opus and Sonnet: Anthropic Unveils Its Latest AI Innovations

This Thursday, May 22, 2025, in a statement released on its blog, Anthropic unveiled its new series of models, Claude 4, available in two variants: Opus and Sonnet. These latest versions feature significant improvements over Opus 3 and Sonnet 3.7. They are already available in the conversational agent. Here’s an overview.

Are you using AI in your work? Take the BDM survey!

Claude 4: Anthropic’s Latest Model Family

Claude Opus 4 is the more advanced model of the two. Anthropic describes it as “the world’s best coding model, with sustained performance on complex, long-duration tasks and agent workflows”. The company also praises its capabilities on long-duration tasks, its “ability to operate continuously for several hours”, and its memory.

Claude Sonnet 4 is an enhancement of Sonnet 3.7. Lighter than Opus 4, which it does not match in most areas, it is designed to offer “an optimal mix of capabilities and practicality”. Also effective in coding tasks, it is distinguished by its high controllability, allowing the user to direct it more precisely according to their intentions, whether adjusting the tone, level of detail, or the type of solution expected.

In keeping with tradition, Anthropic has released a benchmark aiming to demonstrate the capabilities of the two new models, including a comparison with Sonnet 3.7 as well as OpenAI o3, GPT-4.1, and Gemini 2.5 Pro. Opus 4 and Sonnet 4 indeed stand out in coding tasks, with success rates of 72.5% and 72.7% on SWE-bench, compared to 55% for GPT-4.1. However, in other tested areas such as academic reasoning or mathematics, their performances are good but remain within the same range as those of other major models, including Sonnet 3.7.

Additional Features for Claude

Alongside this new launch, Anthropic has introduced several new features. Users will now benefit from the following:

• Extended tool usage (beta): the models can switch between autonomous thinking and using external tools, such as web searches, to enhance the quality of their responses in complex situations.
• Enhanced capabilities: they can use multiple tools simultaneously, follow given instructions more precisely, and better remember information when they have access to local files, extracting key elements to ensure continuity in exchanges.
• New API possibilities: Anthropic’s API now offers four features for developers: code execution, an MCP connector for automating certain tasks, an API for managing files, and an option to temporarily hold requests (up to an hour) to optimize performance.

How to Access Claude 4 Models

The two new models are already available in Claude. However, while Claude Sonnet 4 is accessible to all, a subscription to the Pro plan is necessary to use Claude Opus 4. To activate one of the models, the process is simple:

• Visit Claude’s interface,
• At the bottom right of the command bar, click on the model selector,
• Choose Opus 4 or Sonnet 4.

The previous models (Sonnet 3.7, Opus 3, and Haiku 3.5) are still accessible by clicking on More models.

Claude Code Now Broadly Available

In February 2025, Anthropic introduced Claude Code, a feature designed to assist developers in their daily work. Accessible from a terminal, an IDE (such as VS Code or JetBrains), or via an SDK, Claude Code offers code suggestions, automatically corrects errors, and provides contextual help directly in the development environment. It can also interact with projects on GitHub, for instance, to respond to comments or fix integration errors.

Initially reserved for a limited number of users, the tool is now available to everyone.