Claude Sonnet 4 Gets a 1M Token Context Window (Beta): What Changed and How to Use It

Deep Dive: What Actually Changed

Anthropic's release notes from August 12, 2025 state that Claude Sonnet 4 now supports a 1 million token context window, available in beta.

A quick translation for non-technical readers: a "token" is a chunk of text, roughly three-quarters of a word on average. The "context window" is everything Claude can hold in mind at once during a conversation or task — your instructions, the documents you attach, and Claude's own responses. Until now, the standard window was 200,000 tokens (roughly 150,000 words). The new 1 million token window is five times larger — enough to hold something on the order of 750,000 words in a single request.

Who has access right now

This is where the fine print matters. Per Anthropic's release notes, the beta is limited to:

• API organizations in usage tier 4, or
• Organizations with custom rate limits

Usage tiers are how Anthropic groups API customers based on spending and payment history; tier 4 is the highest standard tier. In plain terms: this feature is currently for businesses accessing Claude programmatically through the API, not for individual users of the Claude.ai website or apps. Anthropic has not said in these release notes when or whether it will reach other tiers or the consumer apps.

Why the size matters

Before this change, working with very large document collections meant splitting them into chunks, summarizing pieces separately, and stitching results together — a process where details get lost between chunks. A 1M window means Claude can see the entire body of material at once, so it can catch connections across documents, like a clause in one contract that contradicts a policy in another.

How to Use This in Your Day-to-Day Work

If your organization qualifies (tier 4 API access or custom rate limits), here's a practical path. If you're not sure whether you qualify, ask whoever manages your company's Anthropic API account.

1. Confirm your access. Check with your technical team or account admin whether your API organization is in usage tier 4 or has custom rate limits. Anthropic's documentation on the 1M context window (linked from the release notes) covers the specifics they'll need.
2. Identify your "too big to fit" workflows. Make a list of tasks you've avoided or worked around because the material was too large: full quarterly report packages, complete contract sets, months of meeting transcripts, an entire support ticket archive, or a full policy manual.
3. Start with a cross-document question. The biggest advantage of a large window is comparison across documents. A good first test: load a set of related documents and ask Claude to find inconsistencies, gaps, or items that reference each other.
4. Keep your instructions at the front and specific. Even with a huge window, clear direction helps. Tell Claude exactly what to look for and what format you want the answer in, rather than asking open-ended questions of a massive document pile.
5. Verify outputs on high-stakes work. A larger window doesn't change the need to spot-check. For legal, financial, or compliance material, have Claude cite which document and section each finding came from so a human can confirm it quickly.

If you don't have access yet

You can still prepare. Organize the large document sets you'd want to analyze, and note the questions you'd ask across them. When access broadens, you'll be ready to test immediately. In the meantime, the standard 200,000 token window still handles substantial workloads — a few hundred pages at a time.

Note: claudedetails.com is an independent publication and is not affiliated with or endorsed by Anthropic. The details above come from Anthropic's own release notes dated August 12, 2025.

---

claudedetails.com is an independent publication and is not affiliated with, endorsed by, or sponsored by Anthropic. "Claude" is a trademark of Anthropic, PBC, used here for identification purposes only. Product details can change — always confirm specifics on Anthropic's own site before making decisions based on this post.