Anthropic launches Opus 4.6, with greater performance and new options such as agent teams

Anthropic has launched Opus 4.6, its updated artificial intelligence (AI) model with better coding, reasoning and task management skills, thanks to its million-token context window and new agent teaming features, as well as increased performance.

The technology company launched Opus 4.5 in November of last year, however, it continues to develop its coding model with an update that arrives just two months later, with a view to improve the Claude Code service experience the specialized version of the Claude assistant focused on programming and technical assistance for developers.

In this sense, Opus 4.6 improves the coding skills of its predecessor this is because you plan with more precision can operate more reliably in larger codebases and has better code review and debugging capabilities to “detect their own mistakes.”

This was shared by Anthropic in a statement on its blog, where it detailed that, for the first time in the Opus model family, it includes a 1 million token context window in its beta version. This means you can retrieve a greater amount of information and work with larger databases, as well as process larger documents.

In practice, the capabilities of Opus 4.6 will help with everyday tasks such as performing financial analysis, conduct research, and use and create documentsspreadsheets or other types of related presentations.

Likewise, it is also integrated into Claude Cowork, the tool designed so that any user can easily schedule and manage various tasks in their daily lives, where they can carry out actions autonomously using all their improved skills.

NEW CAPABILITIES IN OPUS 4.6

However, one of the new features that Opus 4.6 integrates is the ability to form what the company has called “teams of agents” in Claude Code to work together. These equipment can split larger tasks into segmented jobs to carry them out more efficiently.

“A session acts as a team leader, coordinating work, assigning tasks and synthesizing results. The Teammates work independently, each in their own context window, and communicate directly with each other“Anthropic explained in this regard, clarifying that this is an experimental function.

According to Anthropic’s product director, Soctt White, this option is similar to having a human team working on specific tasks, which can be coordinated in parallel to work faster.

Following this line, Opus 4.6 also includes a new “compaction” capacity to summarize your own context and execute longer tasks. Additionally, with the new function of adaptive thinking, the model can “pick up contextual cues about how much to use its extended thinking,” to deliver good results more efficiently, with more control over the speed and cost of use.

In addition to all this, Claude Opus 4.6 is now available on PowerPoint in preview to help generate complete presentations from prompts.

HIGH PERFORMANCE AND SECURITY

Anthropic has also highlighted the performance of the model, which achieves Highest score in Terminal-Bench 2.0 agentic coding assessment and leads on Humanity’s Last Exam, a complex test that measures multidisciplinary reasoning. Likewise, he has detailed that it surpasses other models in the sector, such as the GPT-5.2 model from OpenAI.

According to the company’s tests, they have verified that “the model focuses more on the more complex parts of a task without being promptedmoves quickly through the simpler parts, handles ambiguous problems more judiciously, and stays productive during longer sessions.”

In addition to all this, the technology company has ensured that Opus 4.6 operates safely, with low rates of misaligned behavior in security assessments. With all this, Claude Opus 4.6 is now available on claude.ai, the API and the main cloud platforms for users subscribed to Claude Pro.