Token-Efficient Tool Use ⚠ verify

Definition

A feature that reduces the number of output tokens Claude spends when calling tools. Originally gated behind a beta header, it is now built into the model on Claude 4+.

Key points

• Beta header (exact): token-efficient-tools-2025-02-19.
• Built-in on Claude 4+: now built into all Claude 4+ models — no header needed, and the header has no effect on current models.
• Migration: when migrating from Claude 4.1 or earlier, docs say to remove the legacy headers token-efficient-tools-2025-02-19 and output-128k-2025-02-19.
• Supported models (current docs page): Opus 4.5, Opus 4.6, Opus 4.7, Opus 4.8, Sonnet 4.6, Sonnet 5 (built-in). No documented error modes or disabling edge cases.
• Token-reduction figures: the current platform docs page does not state a percentage. The original launch (Claude 3.7 Sonnet) claimed up to 70% reduction in output tokens, ~14% average — these are legacy 3.7-era numbers, not on the current page. ⚠ verify (2026-07-08).

Why it matters for the exam

• Tests whether you know the feature is now built-in (header no longer required) vs. the legacy beta-header workflow.
• The exact beta header string and the "remove on migration" guidance are precise, testable details.

Common gotchas

• Quoting the 70% / 14% figures as current — they are legacy Claude 3.7 Sonnet numbers and absent from today's docs. ⚠ verify.
• Assuming you still need the beta header on Claude 4+ — it is a no-op there.
• Leaving token-efficient-tools-2025-02-19 / output-128k-2025-02-19 headers in place after migrating from ≤4.1.

See also

• Effective Tool Descriptions
• Structured Outputs

Sources

• docs-deep-api