Anthropic Introduces Prompt Caching for Claude 3.5 Sonnet: 90% Cost Reduction
Anthropic launched native prompt caching across Claude API endpoints, allowing developers to cache long system prompts and documents at a 90% discount on cached input tokens.
Verified State Diff
Comparison Mode:
- Previous State
Full input token price charged on every repetitive API call ($3.00/MTok).
+ Verified New State
Cached prompt tokens priced at $0.30/MTok (90% discount) with 2x lower TTFT latency.
Impact & Verification Analysis
WHO IS AFFECTED
AI engineers building conversational agents, coding assistants, and multi-turn workflows.
WHY IT MATTERS
Massively reduces operating costs for RAG systems and autonomous agent loops.
Full Fact Overview
Prompt caching stores frequent context across requests for up to 5 minutes, significantly lowering latency and cost for large codebase indexing and document chat pipelines.
Multi-Source Evidence Chain (1)
TRACKED ENTITY
Explore all historical Anthropic changes