A million tokens. That’s roughly 750,000 words — the equivalent of feeding Claude the entire Lord of the Rings trilogy, your company’s full legal history, and three years of Slack messages, and still having room left over. As of early 2026, both Claude Opus 4.6 and Sonnet 4.6 ship with a 1M token context window, and the gap between “AI assistant” and “AI that actually knows your entire codebase” has quietly collapsed. This isn’t a theoretical capability sitting behind a waitlist. Anthropic is approaching $19B in annualized revenue, has launched self-serve Enterprise plans you can activate without talking to a salesperson, and just announced a $100M Partner Network on March 12, 2026. The infrastructure is live. The question is whether you know what to do with it.
What Actually Changed Between 4.5 and 4.6
The version bump sounds minor. It isn’t. Here’s what’s different in practice across both models.
Claude Opus 4.6 is Anthropic’s most capable model, full stop. The 1M token context window is the headline, but the more interesting signal came from Anthropic’s own Frontier Red Team: they used Opus 4.6 to find over 500 vulnerabilities in production open-source code. That’s not a marketing demo — that’s Anthropic’s internal security researchers using the model as an actual tool and reporting what it found. On legal, financial, and coding benchmarks, Opus 4.6 outperforms its competitors in the current landscape. If you’re on Max, Team, or Enterprise plans, you get 1M context for Opus 4.6 by default in Claude Code.
Claude Sonnet 4.6 launched February 17, 2026, at the same price as 4.5 — which matters a lot. Better performance, same cost, with the 1M token context window available in beta. It also brings improved agentic search capabilities and uses fewer tokens to accomplish equivalent tasks, which directly affects API costs at scale. Anthropic has also deprecated Opus 4 and 4.1 from the model selector entirely, so the path forward is clear whether you’re building or just using the chat interface.
The practical upshot: Sonnet 4.6 is now the obvious default for most API use cases where you were previously making a cost/performance tradeoff. Opus 4.6 is for the jobs where you genuinely need maximum capability — deep code review, complex legal analysis, multi-document financial synthesis — and you’re willing to pay for it. For a broader look at how Claude by Anthropic has built its reputation for high-stakes tasks, that context is worth having before you commit to a plan.
The 1M Context Window in Practice: What You Can Actually Do With It
Context windows are one of those specs that sound impressive and feel abstract until you run into the wall of a smaller one. Here’s what 1M tokens actually unlocks across different types of work.
For Developers
You can load an entire large codebase into context and ask Opus 4.6 to audit it for security vulnerabilities — which is exactly what Anthropic’s red team did to find 500+ issues in production open-source code. You can also use Claude Code with the full 1M window by default on Max/Team/Enterprise plans. The Skills API (Agent Skills) lets you organize capability into structured folders with SKILL.md files, and there are pre-built skills for PPTX, XLSX, DOCX, and PDF generation. Claude Code is now included as standard in every Team plan seat, not an add-on.
For Legal and Financial Teams
Loading a full contract history, a merger and acquisition data room, or years of financial filings into a single context is now operationally feasible. Opus 4.6 outperforms on legal and financial benchmarks specifically, and Claude Cowork’s domain-specific plugins for legal and financial analysis give you an interface built around those workflows rather than a generic chat box. If you’re evaluating how Claude stacks up against other platforms for these use cases, every major AI platform in 2026 is worth a look before committing to a stack.
For Knowledge Workers Generally
The framing Scott White, Anthropic’s Head of Product for Enterprise, used is “vibe working” — knowledge workers directing AI instead of writing code themselves. The Anthropic engineering team reportedly uses Claude for roughly 60% of their work and ships 60 to 100 internal releases per day. That’s the internal benchmark they’re building toward for external users.
Claude Cowork: The Desktop App Nobody Is Talking About Enough
Claude Cowork launched in research preview at the end of January 2026, and it represents a meaningfully different interface model from the web chat. It’s a macOS desktop app (macOS first) that runs in an isolated VM on your local computer. That means it has full access to your local files and MCP integrations — not a sandboxed version, not a file upload dialog, but actual filesystem access within a controlled environment. For a thorough breakdown of how it works and what to realistically expect from it, the Claude Cowork deep dive covers the research preview in detail.
What’s notable about Cowork structurally: it was built using Claude Code in 10 days. Anthropic is eating its own cooking in a visible way. The persistent agent thread for Pro and Max plan users works across both mobile and desktop, so a task you start on your phone can continue on your desktop without losing context.
The domain-specific plugins cover legal, financial analysis, HR, engineering, and operations. These aren’t cosmetic themes — they’re designed to surface the right tools and context for work in those domains without requiring the user to understand what MCP integrations are or how to configure them. The “vibe working” framing is actually useful here: the interface is designed so that a senior lawyer or a CFO can direct complex AI workflows without needing to understand what’s happening under the hood.
One honest caveat: Cowork is still in research preview. Features and availability may shift, and the plugin ecosystem is early. If you need production stability, the standard Claude.ai interface plus Claude Code is still the more mature stack.
Claude Code in 2026: What’s Actually Shipping
Claude Code has moved from “interesting developer tool” to “daily release cadence, included in every Team plan.” Here’s a snapshot of what’s live or in research preview as of March 2026. For a fuller walkthrough of the feature set and how to get the most out of it, how to use Claude Code in 2026 covers the current state in detail:
- Skills API (Agent Skills): Organized folders with SKILL.md files define reusable capabilities. Pre-built skills for PPTX, XLSX, DOCX, and PDF are available out of the box.
- –channels permission relay: In research preview. Allows more granular control over what Claude Code can access and do during agentic workflows.
- –bare flag: Designed for scripted automation — strips the interactive interface for use in pipelines.
- Voice mode: You can speak to Claude Code rather than type. Useful when you’re context-switching between tools