Anthropic launches Claude Opus 4.6 featuring new ‘agent teams’ capability

After spinning the global tech markets earlier this week, Anthropic is back in the spotlight with another major release.

The company has unveiled Claude Opus 4.6, a new version of its flagship AI model that packs more coding intelligence, longer reasoning memory, and a first-of-its-kind ‘agent teams’ feature designed to make AI collaboration more powerful than ever.

New Engineering blog: We tasked Opus 4.6 using agent teams to build a C compiler. Then we (mostly) walked away. Two weeks later, it worked on the Linux kernel.

Here's what it taught us about the future of autonomous software development.

Read more: https://t.co/htX0wl4wIf pic.twitter.com/N2e9t5Z6Rm

— Anthropic (@AnthropicAI) February 5, 2026

The timing couldn’t be more dramatic. Anthropic’s earlier product rollout, including a legal AI tool that spooked investors, triggered a global tech selloff wiping out nearly a trillion dollars in market value. Now, the same company behind that market chaos is upping the ante with a model that pushes the limits of what AI systems can do autonomously.

Anthropic launches Claude Opus 4.6: What is it?

Claude Opus 4.6’s headline feature is its 1-million-token context window, currently in beta. It means the AI can now “remember” and process massive amounts of information at once, roughly equivalent to hundreds of thousands of words. Instead of splitting large documents, codebases, or datasets into parts, Opus 4.6 can handle them in a single, continuous thought process.

Tokens, the small pieces of data AI models use to understand text, serve as the system’s mental units. With a million of them in play, Opus 4.6 effectively gains a supercharged working memory, giving it the ability to conduct complex, multi-step reasoning and long-form planning with far greater coherence.

Anthropic says the new version is particularly strong in coding, debugging, and software development. It can detect and correct its own mistakes, review extensive codebases in parallel, and sustain long-running engineering sessions, all areas where AI reliability often breaks down.

Quick Reads

View All

OpenAI says it blocked China-linked bid to smear Japan’s Prime Minister using ChatGPT

US calls robot makers to the table as China races ahead in humanoid revolution

Benchmarks back that up. Opus 4.6 achieved top marks in the Terminal-Bench 2.0 coding evaluation and outperformed OpenAI’s GPT-5.2 in finance, legal, and research domains. It also led on BrowseComp, a test of an AI model’s ability to find difficult-to-locate information online.

A step closer to AGI

The standout new addition is the ‘agent teams’ capability, allowing multiple instances of Claude to work together on complex tasks. Think of it as AI collaboration at scale, one model can write code while another reviews it, a third runs tests, and a fourth compiles documentation, all in sync.

This functionality builds on Claude Cowork, Anthropic’s environment for autonomous task execution. Inside Cowork, Opus 4.6 can perform extended work, from drafting presentations and spreadsheets to conducting financial analyses, with minimal human supervision.

On the safety front, Anthropic claims that Opus 4.6 shows fewer unnecessary refusals and reduced misaligned behaviour like deception or hallucination. It also introduces new cybersecurity assessments that evaluate both defensive and offensive capabilities, ensuring the model operates within safe and predictable limits.

Beyond performance, Opus 4.6 reflects Anthropic’s growing push toward artificial general intelligence (AGI) — a future where machines could handle virtually all forms of intellectual work.