Source

Claude Models Overview — Anthropic docs

Finding

Claude Opus 4.6 and Sonnet 4.6 support a 1M token context window via beta header anthropic-beta: context-1m-2025-08-07. Long context pricing applies above 200K tokens. Useful for large codebase analysis, document processing, and long-running agent sessions.

Applicability

Zeph's Claude provider (crates/zeph-llm/src/claude.rs) currently uses the standard 200K context. Adding an opt-in config option allows users to enable the extended window for code indexing and multi-session workloads:

[llm.claude]
enable_extended_context = true  # sends context-1m-2025-08-07 beta header

The zeph-index use case (large codebase repo map injection) and graph memory recall are the primary beneficiaries. Long context pricing should be documented in the config.

Note: Zeph's own compaction pipeline should not fire before the extended limit is reached when this beta is enabled — context_budget_tokens in config should be respected above 200K.

Priority

Medium — immediately useful for zeph-index and large codebase workflows; pricing increase is opt-in.