Feature: Long-context (≈1M) support for Claude Sonnet 4 in Claude Code

Summary

Please add first-class support in Claude Code (CLI / IDE integrations) for Claude Sonnet 4’s 1M-token context window and expose a way to pass Anthropic beta headers—specifically context-1m-2025-08-07. This would let the tool operate on entire repos / large docs without manual chunking.
– Anthropic docs: 1M context + required beta header.
Anthropic

– Availability: Anthropic API + Amazon Bedrock (Vertex AI “coming soon”).
Anthropic

Problem

There isn’t a straightforward, tool-level way to enable long-context when it’s available on the account/provider. Whole-repo refactors and cross-file reasoning require dropping down to raw API calls or ad-hoc configs.

Proposal

• Provide a simple toggle/setting to enable long-context mode when supported by the account/provider.
• Abstract provider-specific opt-ins behind Claude Code’s config/UI.
• Document expected limits, pricing/rate-limit considerations, and failure modes.
• Gracefully handle ineligible accounts with clear, actionable messaging instead of hard errors.
• Optional: when prompts approach the default context, nudge users to enable long-context (non-blocking).

Why this helps

Enables whole-repo operations, large-scale refactors, and multi-doc synthesis—the core Claude Code use cases—without extra glue code.