Claude Code now defaults to a 1 million token context window on Opus 4.6 and Sonnet 4.6, with no opt-in required and no pricing surcharge. This gives developers 5-10x more working memory than Cursor, Copilot, or Windsurf, and it changes how long coding sessions work.

TL;DR

• What changed: On March 13, 2026, Anthropic made the 1M context window generally available (GA) for Opus 4.6 and Sonnet 4.6. Previously it was 200K tokens by default and required beta headers or extra usage credits to go beyond that.
• Pricing: The long-context surcharge (2x input, 1.5x output for tokens beyond 200K) is gone. Flat rate pricing now applies regardless of context length.
• Who gets it: Max, Team, and Enterprise plan users on Claude Code. No extra purchase needed.
• Practical impact: 15% fewer compaction events, ~75,000 lines of code in a single session, 600 images/PDFs per request (up from 100).
• The catch: A bigger window does not mean you should ignore context management. Use /compact at 40% usage, and always maintain a solid CLAUDE.md file.

What Actually Changed on March 13?

Anthropic moved the 1 million token context window from beta to general availability across three layers simultaneously. The API no longer requires the context-1m-2025-08-07 beta header or Tier 4 status. The pricing surcharge for tokens beyond 200K is eliminated entirely. And Claude Code now defaults to the full 1M window for Max, Team, and Enterprise users without needing "extra usage" credits.

This was the final step in a gradual rollout. Claude Code v2.1.50 first gave Opus 4.6 fast mode access to 1M tokens. Version 2.1.73 made Opus 4.6 the default model on Bedrock, Vertex, and Microsoft Foundry. And v2.1.75 (March 13) removed the last gate: the extra usage requirement.

If you want to opt out and stick with the 200K window, set the environment variable CLAUDE_CODE_DISABLE_1M_CONTEXT=true.

Sources: Anthropic 1M Context GA Blog Post | Claude Code Changelog

How Does This Compare to Other AI Coding Tools?

The 1 million token context window is the largest default among dedicated coding assistants. Here is how it stacks up against the competition.

ToolEffective Context WindowNotesClaude Code (Opus 4.6)1,000,000 tokensDefault on Max/Team/Enterprise. No surcharge.Cursor~120K-200K tokensSupports models up to 200K but effective usable context is lowerGitHub Copilot~128K tokensDraws from open files, recent files, repo structureWindsurf~100K tokensSession-level context tracking with codebase awarenessGemini (via API)1M-2M tokensHas had 1M+ for longer, but variable quality at rangeGPT-5.4 (OpenAI)1,000,000 tokensRecently launched with 1M, but loses 54% retrieval accuracy at scale

Raw numbers only tell part of the story. Opus 4.6 scores 78.3% on MRCR v2 at 1M tokens, which Anthropic says is the highest among frontier models. GPT-5.4 reportedly loses 54% of its retrieval accuracy scaling from 256K to 1M. In other words: it is not just about how much you can fit in the window. It is about whether the model can actually use what is in there.

IDE-based tools like Cursor and Copilot compensate with semantic indexing and vector search over repositories. That approach can be more efficient for targeted lookups. But for long autonomous sessions (multi-hour CI loops, large refactors, complex debugging chains), raw context capacity matters. Having the model remember your decisions from 30 minutes ago without needing to re-explain them is a real workflow improvement.

What Does 1 Million Tokens Actually Look Like in Practice?

One million tokens translates to roughly 75,000 lines of code, or hundreds of documents loaded into a single session. Anthropic's own data shows a 15% decrease in compaction events across real Claude Code usage since the change.

That means fewer moments where the model suddenly "forgets" the architecture decision you made at the start of the session. Fewer times where you need to re-explain what you are building. And for media-heavy workflows, the limit jumped from 100 to 600 images or PDF pages per request.

Practically, this benefits three types of workflows the most:

1. Long autonomous agent runs where Claude Code iterates on CI failures, runs tests, and fixes issues across multiple files over extended periods
2. Large codebase refactors where you need the model to hold awareness of how changes in one file affect dozens of others
3. Multi-system debugging where you are jumping between logs, config files, test output, and source code in a single session

For most daily tasks (writing a function, fixing a bug, generating a component), you will not come close to 1M tokens. Many developers report staying under 100K in typical sessions. The value is in removing the ceiling so you never hit it during the sessions that matter most.

Related reading: How to build a 99% SEO website in 12 minutes with Claude Code | Claude Code memory for marketing and SEO

Why You Still Need to Manage Your Context Window

A 1 million token context window does not mean you should treat it like an unlimited buffer. This is the part most people will get wrong.

Here is the reality: even with 1M tokens available, model performance can degrade as context fills up. More tokens means more noise for the model to sort through. Important instructions from early in the session can get diluted by thousands of lines of tool output, file reads, and intermediate steps. The model is not losing the information. It is just competing with more information for attention.

The 40% rule: When your context window reaches approximately 40% usage, run the /compact command. This is not a panic button. It is routine maintenance. When you run /compact, give it clear instructions on what to preserve. Tell it what decisions were made, what files were modified, and what the next steps are. A good compact instruction looks like this:

/compact Preserve: we are refactoring the auth middleware in src/auth/. Files modified so far: middleware.ts, session.ts, types.ts. Decision: using JWT with refresh tokens instead of session cookies. Next: update the route handlers in src/routes/ to use the new middleware.

Without those instructions, compaction will summarize generically, and you will lose the specifics that matter.

The CLAUDE.md advantage: No matter how large your context window is, a well-maintained CLAUDE.md file will always outperform relying on context alone. Your CLAUDE.md loads at the start of every session and every compaction. It is the one thing that persists no matter what.

A good CLAUDE.md contains:

• Project structure and key file locations
• Coding conventions and patterns used in the project
• Common commands (test, build, deploy)
• Architecture decisions and their rationale
• Things the model should never do (destructive commands, specific patterns to avoid)

Think of it this way: the 1M context window is your short-term memory. CLAUDE.md is your long-term memory. You need both. The context window handles the current session. CLAUDE.md handles everything that should survive across sessions.

Related reading: Build the perfect SEO copywriter with Claude Skills | Replace Zapier and n8n with Claude Code cron jobs

What Does the Pricing Change Mean for Your Budget?

The pricing change might be more significant than the context window itself. During the beta period, using more than 200K tokens meant paying 2x on input tokens and 1.5x on output tokens. For heavy users running long agent sessions, this added up fast.

Now Opus 4.6 stays at $5 per million input tokens and $25 per million output tokens, regardless of whether you use 50K or 950K of the window. Sonnet 4.6 stays at $3/$15. A 900K-token request costs the same per-token rate as a 9K-token request.

For Claude Code Max/Team/Enterprise subscribers, this is even simpler: the 1M window is included in your plan. No extra usage credits needed. You pay your subscription and use the full window.

The Cursor community immediately noticed this change and asked when it would cascade to their pricing, since Cursor pays Anthropic's API rates on behalf of users.

Who Should Care About This Update?

This update matters most if you fall into one of these categories. If you use Claude Code for anything beyond quick one-off tasks, the larger context window removes friction you may not have even noticed.

• Solo developers working on full-stack projects where you jump between frontend, backend, database, and deployment config in a single session
• Teams using Claude Code for code review where the model needs to hold the full PR diff plus surrounding context
• Anyone running autonomous agent workflows (Claude Code with cron jobs, CI pipelines, or monitoring scripts)
• Content creators and marketers who use Claude Code for SEO automation, batch content generation, or Google Workspace integrations
• API developers who previously had to manage the beta header and Tier 4 requirement

If you are already on Max, Team, or Enterprise, you do not need to do anything. The 1M window is already active. Check it by looking at the model identifier in your Claude Code status line. It should show "Opus 4.6 (1M context)."

Related reading: How to connect Claude to SEO data | What is Model Context Protocol

FAQ

Does the 1M context window cost extra on Claude Code?

No. As of March 13, 2026, the 1 million token context window is included by default for Max, Team, and Enterprise plan users. The previous "extra usage" requirement and the long-context pricing surcharge have both been removed. You pay your normal subscription rate.

Should I let my context window fill up to 1M tokens before compacting?

No. Run /compact when you reach approximately 40% of your context window. Larger context means more noise for the model to sort through, which can reduce the quality of responses. When compacting, provide specific instructions about what to preserve: files modified, decisions made, and next steps. This keeps your session focused and productive.

What is a CLAUDE.md file and why does it matter with a larger context window?

A CLAUDE.md file is a markdown file in your project root that Claude Code reads at the start of every session. It contains project structure, coding conventions, key commands, and architecture decisions. Even with 1M tokens of context, CLAUDE.md acts as persistent long-term memory that survives compaction and session restarts. The context window is short-term memory. CLAUDE.md is long-term memory. You need both.

How does Claude Code's 1M context compare to Cursor or GitHub Copilot?

Claude Code's 1M token context window is roughly 5-10x larger than Cursor (~120-200K), GitHub Copilot (~128K), and Windsurf (~100K). More importantly, Opus 4.6 maintains 78.3% retrieval accuracy at 1M tokens, which is the highest among current frontier models. Competing tools compensate with semantic indexing and vector search, which works well for targeted lookups but not for maintaining session-long awareness during complex multi-file tasks.

Can I opt out of the 1M context window and use the old 200K default?

Yes. Set the environment variable CLAUDE_CODE_DISABLE_1M_CONTEXT=true to revert to the 200K context window. Some developers prefer this for faster response times or tighter cost control on API usage. The opt-out is per-session, so you can switch between them as needed.

Bottom Line

The 1M context window going default is a meaningful upgrade, not because most sessions need a million tokens, but because it removes the ceiling for the sessions that do. Combined with the pricing surcharge elimination, it makes Claude Code significantly more practical for long, complex coding sessions.

But the real takeaway is this: context window size is a tool, not a strategy. The developers who get the most out of Claude Code are not the ones with the biggest context windows. They are the ones who maintain clean CLAUDE.md files, compact proactively at 40%, and structure their sessions with clear intent.

A well-organized 200K session will outperform a chaotic 1M session every time. The 1M window just means that when you do need the space, it is there.

Want to learn how to use AI tools like Claude Code for SEO and content? Join the AI Ranking community where we teach practical AI-powered SEO workflows every week.

Sources: Anthropic 1M Context GA Announcement | Claude Code Changelog v2.1.75 | Anthropic API Docs: Context Windows | The Decoder: Anthropic Drops Surcharge | Simon Willison's Coverage

TL;DR

Most websites dump all their services on one page and wonder why they can't rank. The fix is simple: one page per service, one search intent per page. This guide covers the 3-step structure, the 50% content differentiation rule, schema markup, and how to scale across multiple locations without getting flagged for duplication.

30 sec readSkip to full article below

Why Do Most Websites Still Struggle to Rank on Google and AI Search?

Because their website structure is broken. It's not backlinks, it's not content volume, and it's not Google's algorithm randomly hating your site. After reviewing thousands of websites over 10+ years in SEO, the pattern is clear: if your website isn't properly organized, search engines can't understand what each page is actually about.

And this matters now more than ever. AI-referred website sessions grew 527% in recent months, and AI search visitors convert 4.4x better than organic search visitors. But AI search engines like ChatGPT, Perplexity, and Google AI Overviews only cite 2 to 7 domains per response. If your site structure is a mess, you're not making that shortlist.

The good news? Fixing your structure is one of the highest-impact things you can do. Let me show you exactly how.

What Is the Number One SEO Mistake Killing Your Rankings?

Putting all your services on one page.

It might look something like this: you have a page called "Our Services" and you list everything you offer. Emergency plumbing, pipe leak repair, kitchen sink installation, all crammed together. You think you're being efficient. You're actually shooting yourself in the foot.

Here's why. When someone searches for "emergency plumbing," Google needs to find the best page that answers that specific search intent. Which page do you think it's going to choose? A page that mentions emergency plumbing alongside five other services? Or a dedicated page focused entirely on emergency plumbing?

The dedicated page wins every time.

Google doesn't rank websites. It ranks individual pages. One page, one service, one search intent. That's the rule.

And this isn't just a Google thing. It's even more critical for AI search engines. The more specific your pages are, the more likely you are to get cited. Remember: AI responses only reference a handful of sources, and pages with answer capsule structures see 40% higher citation rates.

How Should You Structure Your Service Pages? (3 Steps)

Break every service into its own dedicated page, organize them hierarchically, and connect them with internal links. Here are the three steps in detail.

Step 1: Identify Every Service and Give It a Page

List every service you offer (or plan to offer) and create a dedicated page for each one. Even if that's 10, 20, or 30 services, each one gets its own page.

Why? Because each page targets a specific keyword solving a specific problem your customer has. A single "services" page can't rank for 30 different keywords. Thirty dedicated pages can.

Step 2: Organize Pages Into a Logical Hierarchy

You still have one main "Services" parent page that lists everything. But each listing links to the dedicated service page underneath it. Your URL structure should follow the same hierarchy:

• /services/emergency-plumbing
• /services/kitchen-sink-installation
• /services/pipe-leak-repair

This gives Google a clear signal about how your services relate to each other and builds topical authority across your entire site.

Step 3: Internally Link Between Related Services

Link from your emergency plumbing page to kitchen sink installation. Link from that page to pipe leak repair. And back again.

Internal linking is a critical component of SEO because it helps with three things: user navigation, faster indexing, and showing Google the overall structure and relationships across your site. Schema markup makes this even more powerful, with pages using structured data being 36% more likely to appear in AI summaries.

How Different Does Each Service Page Need to Be? (The 50% Rule)

At least 50% different from every other service page. If your pages look practically the same with just the service name swapped out, Google is smart enough to flag them for content duplication.

And in AI search, this is even more brutal. AI engines cite only one page from a group of "near-duplicates." That means 50 templated pages with the same content equals 49 wasted pages.

Here's what makes each page unique (it goes way beyond rewriting headlines):

SEO Fundamentals:

• Unique title tag with the service name, benefit, qualifier, and location
• Unique meta description
• Clean URL structure
• Different H1 for each page

Content Sections That Differentiate:

• Service introduction: How does this specific service solve a specific problem?
• How it works: Walk through the process for this particular service
• FAQs: Frequently asked questions specific to that service (bonus: FAQ schema gives you 3.2x higher citation probability in AI search)
• Pricing factors: What affects the cost of this particular service?
• Urgency-matched CTAs: Emergency plumbing gets "Call Now." Pipe leak detection gets "Book an Inspection." Small difference, big impact
• Service-specific reviews: Customer testimonials that mention that exact service
• Unique images: AI image generators can create realistic photos of you or your team performing each service

Steven, one of our AI Ranking community members, took this approach with over 800 location pages. He's now getting 105 appointments per month, with pages indexing in under an hour. The structure is what makes it work.

Why Does Service Schema Matter for AI Search?

Service schema is a small piece of code in the header of each service page that tells AI search engines exactly what's on that page, instantly and unambiguously.

Think of it as a translation layer. Structured data improves GPT-4's accuracy from 16% to 54% when processing page content, and pages with schema markup are 36% more likely to appear in AI summaries.

Each service page should have:

• Service schema describing what the service is, who provides it, and the service area
• FAQ schema for the frequently asked questions section
• Local business schema if you serve specific geographic areas

This is one of the easiest wins in SEO. You write the schema once per page, and it keeps working for you every time an AI search engine crawls your site.

How Do You Scale Service Pages Across Multiple Locations?

Use the same structure, but add a location layer on top.

If you're a plumber serving London, Manchester, and Birmingham, you need a location parent page that lists all your service areas. Each location then becomes its own parent page linking to the services offered there:

• /locations/london/emergency-plumbing
• /locations/london/pipe-leak-repair
• /locations/manchester/emergency-plumbing

The 50% differentiation rule still applies. Here's how to make location pages unique without it being a nightmare:

• Embed a Google Map of that specific location (just search the area in Google Maps, click Share, then Embed, and paste the HTML at the bottom of the page)
• Area-specific details: Mention a local road, church, school, or landmark where it makes sense
• Location-specific schema: Local business schema that specifies the service area
• Location tags: Help Google understand which content belongs to which area

For local businesses, content quality beats proximity in AI search. AI Overviews have zero distance correlation, unlike the traditional Local Pack. That means a well-structured location page can outrank closer competitors who have sloppy site architecture.

How Do You Generate Unique Images for Every Service Page?

Use AI image generators. You don't need a photographer following you around.

Upload a photo of yourself (or your team) to an AI image generator like Google AI Studio and use a prompt that places you in the context of performing that specific service. For example: "A plumber fixing a leaky pipe underneath a kitchen sink, professional setting, natural lighting."

The result is a unique, realistic-looking photo for each service page. And here's the SEO bonus: you can add descriptive alt text to each image (something like "plumber in London fixing a leaky pipe"), which gives you another differentiation signal that both Google and AI search engines can pick up on.

These small 1% improvements compound. A unique image with a descriptive alt tag, combined with unique content, unique schema, and a unique CTA adds up to a page that is genuinely different from every other service page on your site.

Frequently Asked Questions

How many service pages should I create?

As many as you have distinct services. If you offer 15 services, you need 15 pages. Each one targets a specific keyword and solves a specific problem. Don't hold back because you think "too many pages" is a thing. Google and AI search engines prefer specificity.

Do I really need 50% unique content on every page?

Yes. While there's no official number from Google, SEO professionals consistently find that 40-50% is the minimum unique content threshold before pages start getting grouped as duplicates. Aim for 50% or higher to be safe.

Can I use AI to generate the content for service pages?

Absolutely. The key is making sure the AI generates genuinely different content for each page, not just swapping the service name. Use a prompt that focuses on the specific problem each service solves, the process, FAQs, and pricing factors. If you generate all pages with the same AI tool using the right prompts, it will naturally produce differentiated content.

What if I only serve one location?

You still need individual service pages. Skip the location layer and focus on making each service page as strong and differentiated as possible. Add your single location to the schema and title tags.

Ready to Fix Your Website Structure?

If you want the free service page checklist I mentioned (plus image generation prompts and a page content generation prompt), grab the AI Search Starter Kit. Just drop your email and I'll send everything over.

Inside the kit, you'll find:

• The complete service page checklist
• AI image generation prompts to create unique service photos
• A page generation prompt that ensures every service page is properly differentiated
• Stats and research sources to back up your SEO strategy

And if you want to go deeper, the AI Ranking community has weekly Q&As, a full course library, and 477+ members sharing what's working right now in AI search. William Moon, a financial advisor in Arizona, used these same structural principles to take his CTR from 0.3% to 2.3% and close a $165,000 deal from organic search.

Your website structure is the foundation everything else builds on. Get it right, and everything from content to schema to AI citations starts working in your favor.