Claude 1M Context: What No One Tells You..

Anthropic's 1M-token context on Opus 4.6 and Sonnet 4.6 ships at the same per-token price as the 200K window — $5/$25 and $3/$15 per million input/output, with no rate-limit penalty and 600 images/PDF pages per request (~830K usable input tokens).

The wins worth taking

Fewer compactions: Anthropic reports a 15% drop in compaction events, so agent sessions hold state longer. Opus 4.6 also scores 78.3% on MRCR v2 — the highest among frontier models at this length, meaning the model actually *uses* what's inside the window, not just fits it. For full-repo analysis or contract review, you can skip the RAG pipeline (chunking, embeddings, retrieval) and just load the files.

The costs people miss

Latency scales with token count — loading 800K for a quick question makes chat/IDE workflows feel sluggish. Then there's the lost in the middle problem: pages buried in a 500-page dump get under-weighted vs. the first and last sections. Fill the window with noisy files and you get attention dilution, where the model hunts for signal.

The real bill

Pricing is linear. One 1M-token Opus input call is $5; twenty in an agentic loop is $100 in input alone, before output. There's also a soft cost: when the window fits everything, you stop curating, and a carefully constructed 50K token prompt will beat a sloppy 1M dump.

How to use it

Frontload and backload critical instructions, use CLAUDE.md / XML tags as a table of contents, and start at the smallest effective context — only scale up when output quality actually demands it. For email, brainstorming, and Q&A, 200K is still plenty.