Anthropic’s Claude Sonnet 4.6 Launches With Smarter Computer Control and Stronger AI Security

Anthropic says its newest model, Claude Sonnet 4.6, is designed to be better at using computers the way people do — not just answering questions, but completing tasks that unfold across multiple screens, tabs, and steps. The release is scheduled to roll out on Tuesday, and it becomes the default option for people using Claude for free as well as those on the Pro plan.

The direction is clear: AI assistants are being trained to move from “tell me what to do” into “do it with me” and increasingly “do it for me.” That shift is where real productivity gains live — and where real risk starts to rise, too.

What Sonnet 4.6 is meant to do on a computer

Anthropic frames Sonnet 4.6 as a step forward for “computer use,” meaning the model can take actions that normally require a human hand on a keyboard and mouse. In practical terms, the company says the model can handle tasks like:

• Filling out web forms that require multiple steps and checks.
• Coordinating details across several browser tabs without losing track of what matters.
• Working through “office” style workflows that combine reading, copying, and updating information.

Anthropic also acknowledges a reality check: Sonnet 4.6 still isn’t at the level of the most skilled humans at complex computer use. But the company argues the pace of improvement is accelerating — and the new default model is meant to bring those gains to far more people, not just premium users.

A familiar arms race: Anthropic, OpenAI, Google

This release lands in the middle of a wider sprint by major AI labs to build models that can control computers to complete mundane tasks at a user’s command. Anthropic introduced a computer-use option in late 2024, and since then rivals have showcased their own agents aimed at day-to-day work.

The competitive message isn’t only about raw intelligence — it’s about reliability in messy, real environments: pop-ups, login steps, page layouts that change, and workflows that require remembering what you saw two tabs ago.

Coding upgrades without forcing users into premium tiers

Alongside computer use, Anthropic is emphasizing coding — a long-standing focus area for the company. Sonnet 4.6 is positioned as more reliable than its predecessor for programming work, with improvements aimed at how the model handles large or complicated code contexts.

The strategy is also commercial: instead of reserving the best experience for the highest-priced tier, Anthropic is shrinking the gap between “everyday” and “premium” models. Sonnet sits in the middle of Anthropic’s three-model lineup: Haiku for speed and cost, Sonnet as the workhorse, and Opus for harder, deeper jobs like complex reasoning and long-range planning.

Why security matters more when AI can click and act

The most important change that comes with computer-controlling AI isn’t a feature — it’s the risk profile. When an AI can browse, type, and submit, mistakes don’t just sit in a chat window. They can become actions: form submissions, data entry, settings changes, and more.

Anthropic highlights a specific class of threats called prompt injection attacks, where a model is manipulated by malicious instructions hidden in content it reads — for example, a web page that embeds a command designed to override the user’s intent. Anthropic says Sonnet 4.6 is much better than Sonnet 4.5 at resisting those attacks, positioning safety as a core requirement for any agent that can act on a user’s behalf.

The market anxiety behind “agents that do your job”

Anthropic’s expansion beyond its developer base has become a market story of its own. Investors have been jittery about which industries could be disrupted as AI becomes more autonomous — especially if “agentic” models make it cheaper to automate repetitive professional workflows.

That fear has shown up in sharp reactions to recent Anthropic product moves, including automation tools for legal work and models positioned for financial research. The broader theme is that the closer AI gets to completing multi-step tasks reliably, the more it challenges traditional software categories — and even some service jobs built around routine digital workflows.

The growth numbers that frame this rollout

This product shift is backed by a funding and enterprise-growth narrative that has quickly become headline material. In a recent funding round, Anthropic was valued at $380 billion after raising $30 billion. The company has also cited major expansion among high-spend customers, including: 7x year-over-year growth in customers spending over $100,000 annually, and more than 500 customers spending $1 million+ per year.

For Anthropic, the logic is straightforward: if the “default” model can do more real work — faster, cheaper, and with stronger safety — it can widen the user base and then convert more of those users into paid subscriptions and enterprise contracts over time.

For a deeper look at the product announcement straight from Anthropic, you can read the company’s release on Claude Sonnet 4.6.