← Back to the archive Machine
Claude Opus 4.7 — a first look at the 1M context window
Anthropic's Opus 4.7 ships with a one-million-token context window. What changes for real workflows, and what doesn't.
⁂
Anthropic just rolled out Claude Opus 4.7 with a 1M-token context window. The headline is obvious — you can fit a small monorepo into one prompt. The interesting question is what that actually changes day-to-day.
What 1M tokens really buys you
Roughly: ~750k words, or a few hundred files of TypeScript with comments. Things that suddenly become natural:
- Whole-repo refactors where the model can see every caller.
- Long-running coding sessions without manual context pruning.
- Reading a full book or a long incident transcript and querying it.
What it doesn’t fix
- Latency grows with input length. Long prompts are slower.
- Cost is still per-token; full-context prompts add up fast.
- Recall at the deep end of context is good, not perfect — important details still benefit from being near the top.
When to actually reach for it
Use the big window when you genuinely need it: large refactors, document QA, multi-file debugging. For day-to-day coding, smaller focused prompts are still faster and cheaper — and frankly, easier to reason about.
The 1M window is a capability, not a habit.
⁂