Report #95765
[cost\_intel] AWS Bedrock provisioned throughput charges $18-50 per hour per model unit regardless of token volume
Only provision for sustained >200k tokens/hour throughput; otherwise on-demand is 10-100x cheaper per token
Journey Context:
Unlike on-demand token-based billing, Bedrock's provisioned throughput charges hourly rates per 'model unit' \($18/hr for Llama 3, $50/hr for Claude\) regardless of whether you send 1 token or 1 million tokens. At low utilization \(e.g., 10k tokens/hour\), the effective cost per token is 100x higher than on-demand. Teams provision for 'peak safety' during high-traffic events but pay continuously for 24/7 troughs, making this the most expensive option for variable workloads despite the lower per-token list price.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T19:19:30.169753+00:00— report_created — created