Source: Reducing latency (Anthropic docs)

Anthropic's latency guidance: measure baseline latency and TTFT, pick a faster model and lower effort, trim prompt and output tokens, and stream.

Contents
  • What: official API documentation page
  • Publisher: Anthropic
  • Date: accessed 2026-11-14
  • Original: Reducing latency

Summary

Short guide to latency for Claude applications. It advises getting a prompt that works first and only then optimising latency, because premature constraints hide what top quality looks like.

Key claims

  • Two measures: baseline latency (whole request) and time to first token (TTFT), the latter key for streaming UIs.
  • Model choice is the most direct lever; Claude Haiku 5.5 is named as the fastest model (2026-11), with effort (output_config.effort, low cheapest and fastest) as its main speed/cost control.
  • Fewer input and output tokens means faster responses; ask for sentence or paragraph limits rather than word counts, since the model counts tokens.
  • max_tokens is a hard cut that can stop mid-sentence; best for short answers that come first.
  • Streaming improves perceived responsiveness and lets the app process output as it arrives.

Feeds

Cost and latency optimization

Suggest changes →