Claude Code – Disabling telemetry also disables 1-hour prompt cache TTL
anthropic claude
| Source: HN | Original article
Claude Code users discovered a silent regression that links two seemingly unrelated settings: turning off telemetry also disables the platform’s one‑hour prompt‑cache tier. The issue surfaced on Anthropic’s public GitHub, where developers reported that sessions launched with DISABLE_TELEMETRY=1 or CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 no longer benefit from the 60‑minute cache that speeds repeated calls. A related bug shows the same flag unintentionally blocks the Opus 4.6 1M model for Max, Team and Enterprise plans, only restoring it when telemetry is re‑enabled and the cache refreshed.
Why it matters is twofold. First, prompt caching is a core performance optimisation in Claude Code; the default five‑minute window already reduces latency, while the optional one‑hour tier can cut API costs for developers who iterate on the same codebase. Losing that tier forces every request to hit the model anew, inflating response times and cloud spend. Second, many enterprises disable telemetry for compliance or privacy reasons, assuming it merely stops data collection. The bug now creates a trade‑off between privacy and productivity, a dilemma that could push teams back to less efficient tooling.
The glitch arrives as Claude Code gains traction after our April 13 feature on the gamified terminal experience, and it follows a series of recent community‑driven enhancements such as neuro‑symbolic extensions and agent‑as‑a‑service offerings. Anthropic has opened a bug ticket (Issue #45381) and promises a fix, but the timeline remains unclear.
What to watch next: a patch that decouples telemetry from cache eligibility, likely bundled in the next SDK release; clarification on whether the one‑hour cache will become configurable by default; and any policy updates from Anthropic regarding telemetry opt‑outs for enterprise customers. Developers should monitor the repository and consider keeping telemetry enabled temporarily to retain cache benefits until the issue is resolved.
Sources
Back to AIPULSEN