What 1 Million Tokens Actually Means for Business (And Why Most People Are Missing the Point)

On 13 March 2026, Anthropic made 1 million tokens of context generally available in Claude Opus 4.6 and Sonnet 4.6. Most of the coverage focused on the technical specs. Benchmarks. Pricing tiers. Token counts.

That misses the point entirely. Let me explain what this actually means in practice, because I've been using it daily for production work and the implications are bigger than most people realise.

What is a context window, in plain language?

Think of the context window as the AI's working memory. It's everything the AI can "see" and reason about in a single conversation. Previously, most AI tools had a context window of around 128,000 to 200,000 tokens - roughly equivalent to a 300-page book.

1 million tokens is roughly 750,000 words. That's about 3,000 pages. An entire codebase. A full set of legal contracts. Every meeting transcript from the past year. All loaded into a single conversation, all available for the AI to reference simultaneously.

Why this matters more than it sounds

The old way of working with AI on large problems involved a painful dance. You'd load part of your data, get a partial answer, then switch context, reload different data, try to stitch insights together manually. The AI would forget what it saw earlier. You'd spend as much time managing the AI's limitations as you spent on the actual problem.

That friction is gone now.

Here's what my actual workflow looks like. I manage multiple production codebases - web platforms, SaaS products, internal tools. When I need to investigate a bug, I load the entire codebase into a single Claude Code session. Every file. Every component. Every API route. The AI can trace a data flow from the frontend button click, through the API handler, into the database query, back through the response, and into the UI render - all without me having to manually point it at each file.

Last week, I asked it to audit an entire web platform for security issues - roughly 90 files across the codebase. It found 102 bugs in a single pass. Previously that would have required splitting the work across multiple sessions, manually tracking which files had been reviewed, and hoping nothing fell through the cracks between context switches. The lossy summarisation that long-context work previously required is no longer needed.

What this enables for businesses

Legal and compliance. Load an entire contract set - master agreements, amendments, addenda, correspondence - and ask questions that span the full document history. "Does clause 4.2 in the 2024 amendment contradict the liability cap in the original MSA?" Previously, that kind of cross-reference analysis required a paralegal spending days with sticky notes. Now it's a single query.

Codebase analysis. Security audits, architecture reviews, refactoring plans - all performed with full visibility across the entire project. No more reviewing files in isolation and missing the connections between them.

Financial analysis. Load a full year of management reports, board minutes, and financial statements. Ask the AI to identify trends, flag inconsistencies, or draft a board summary that references specific data points from across the entire corpus.

Knowledge management. Organisations that have accumulated years of internal documentation, policies, and procedures can load the lot and get consistent, cross-referenced answers. No more "I think it's in the policy somewhere" followed by thirty minutes of searching.

Customer intelligence. Load every support ticket, NPS response, and product review from the past twelve months. Get pattern analysis that would take a human analyst weeks to produce manually.

The economics are shifting

Here's the part that should get the attention of every business leader: Anthropic is charging standard per-token rates across the full 1M window. A 900,000-token request costs the same per-token rate as a 9,000-token one. There's no multiplier.

That means the cost of comprehensive analysis just dropped dramatically. A full codebase security audit that might cost $15,000-30,000 from a consultancy can now be run as a first pass for a fraction of that. Not as a replacement for human expertise - but as a way to surface the issues that need human attention, rather than paying humans to find them manually.

What businesses should do about this

If you're running a business and wondering where to start, here's my practical advice:

Identify your largest information bottleneck. Where does your team spend the most time manually reviewing, cross-referencing, or synthesising large volumes of information? That's your first use case.

Start with internal analysis, not customer-facing output. Use the extended context for internal reviews, audits, and analysis before you put it in front of customers. Build confidence in the outputs first.

Don't try to boil the ocean. Pick one specific workflow - contract review, codebase audit, financial analysis, customer feedback synthesis - and run a pilot. Measure the time saved against the cost. Then expand.

Invest in someone who understands the tools. The gap between "using ChatGPT occasionally" and "running production AI workflows with 1M context" is enormous. Having someone on your team (or advising your team) who understands how to structure prompts, manage context effectively, and evaluate outputs is the difference between wasted spend and genuine transformation.

The real competitive advantage

The businesses that figure this out first will have an unfair advantage. Not because the technology is secret - it's available to everyone. But because most organisations are still treating AI as a fancy autocomplete, when it's become something fundamentally different: an analytical engine that can hold your entire business context in working memory simultaneously.

The technology just crossed a threshold. The question now is whether your business crosses it too.

If you're trying to figure out where 1M context fits in your operations, get in touch. This is exactly the kind of strategic assessment I do at Cognitiv.