AWS Details Bedrock Prompt Caching: Up to 90% Cheaper Input Tokens on Cache Hits
AWS ML Blog · rss · 2026-09-16
AWS ML Blog published a hands-on guide to Amazon Bedrock prompt caching: prefix-caching repeated context (system prompts, documents, tool definitions) cuts input token costs by up to 90% on cache hits and reduces TTFT, with 75% net savings for repeated-context workloads.
Key points:
- Pricing: cache writes cost 25% more than standard input (2x for 1-hour TTL); cache reads are 90% cheaper. Default TTL is 5 minutes, up to 1 hour on select models.
- Thresholds: Claude Sonnet 4.5/4.6 require 1,024 tokens per checkpoint, Opus 4,096. The Converse API cachePoint syntax works across model families including Claude and Amazon Nova.
- Six progressive scenarios with code: message content caching, system prompt caching, tool definition caching, mixed TTL, tenant isolation, and LangChain integration.
More from Infra
- MLPerf Inference v6.1 results imminent: 30 submitters, new accelerators and platforms — TheKanter · 2026-09-16
- Lambda runs 19 nodes in a 16-node power budget with NVIDIA DSX, +24% throughput — TheZachMueller · 2026-09-16
- NVIDIA unveils DSX AI Factory platform to maximize output per megawatt — nvidia · 2026-09-16
- Claim: run a 125B MoE at 22 tok/s on $320 of used GPUs with llama.cpp — cephaloform · 2026-09-16
- Astera Labs' Leo X-Series memory controllers boost long-context inference TTFT by up to 62% — BenBajarin · 2026-09-16
- Dev measures under 4.5GB VRAM peak throughout full local generation — cocktailpeanut · 2026-09-16