Claude Opus 4.6: Anthropic превращает одну модель в целый ИИ-отдел

Remember the times when we felt like geniuses by adding the phrase "think step by step" to our prompts? Those days seem to be numbered. Anthropic has just moved the goalposts in a game we barely started understanding. The release of Opus 4.6 is not just a routine model update for a couple of benchmark points. It's industry's official recognition: one person is not enough, even if that one is the smartest neural network on the planet. The new concept of "agent teams" changes the very mechanics of how we interact with artificial intelligence.

Let's recall the context. For a long time, Anthropic remained in OpenAI's shadow, playing the role of "that cautious and clever brother." While Sam Altman promised us AGI every other Tuesday, Dario Amodei's team methodically worked on safety and context understanding quality. Claude 3.5 Sonnet already became developers' favorite for its humanity and code cleanliness, but heavyweight Opus needed something more than just a speed boost. It needed a new architecture of meaning. After OpenAI released its o1, focused on deep reasoning, Anthropic decided to go a different way — the way of orchestration.

What exactly changed in Opus 4.6? Instead of trying to digest a complex multi-level task in one huge chunk, the model now knows how to create temporary sub-agents.

Imagine you give a task to a manager, and he himself hires an analyst, copywriter, and editor to complete the project. Opus 4.6 plays that manager role.

It analyzes the request, understands what skills are needed for its implementation, and distributes subtasks between specialized versions of itself. This solves the main problem of large language models — loss of focus when executing long chains of actions. When one neural network tries to be both a strategist and an executor at the same time, it inevitably starts to hallucinate.

Division of labor reduces this risk to almost zero.

Why does this matter right now? We're on the brink of disappointment with ordinary chatbots. Business is tired of "playing questions and answers," business needs results. Opus 4.6 targets exactly this demand. It's a tool for creating autonomous systems that can lead a project from idea to implementation, checking themselves at each stage. If before you built complex chains in LangChain or other frameworks manually, now Anthropic bakes this logic directly into the system core. This is a serious blow to the market of startups that built their business only on wrappers for multi-agent systems.

It's interesting to watch the collision of two philosophies. OpenAI bets on "System 2 thinking" — slow, deliberate reasoning of a single model. Anthropic bets on "collective intelligence" within a single system. Who will turn out to be right will be shown by time, but right now Opus 4.6 looks like a much more practical tool for real tasks, where not only logic matters, but also the volume of work completed. We're transitioning from the era of "AI-assistants" to the era of "AI-employees," and Anthropic has clearly seized the lead here.

The bottom line: Anthropic is turning a neural network from a conversationalist into an operating system for tasks. Will Opus 4.6 be the final nail in the coffin of the prompt engineer profession?