What Amazon Bedrock really costs
200,000 requests a month to Nova Pro cost $0.00 a month, or $0 a year.
aws list prices · checked 17 november 2026 · sources
Bedrock, Nova Pro US East (N. Virginia)
a year $0
you can keep
$0.00 a month
That is what batch inference would save on this model: a month instead, for work that can wait up to a day.
Why the bill is bigger than expected
Output tokens cost several times more than input. Most models charge 3 to 5 times more for each token they write than for each token they read. Long answers drive the bill more than long prompts.
Context is billed on every call. Retrieved documents, chat history and system prompts are input tokens, sent again with every request. Trimming what you send is often the biggest saving.
Batch costs about half. Many models offer batch inference at about 50 % off for work that can wait up to a day: summaries, classification, enrichment.
Sources
- [1]Bedrock on-demand and batch token prices, by model and region. AWS Price List API, AmazonBedrock offer, version 20261111013037, read 2026-11-17.
- [2]The page to check it against. aws.amazon.com/bedrock/pricing, read 2026-11-17.
Models that AWS prices directly; models sold through AWS Marketplace on Bedrock (including the latest Anthropic Claude models) are not listed here. Prices are list prices in USD before tax, credits or private discounts. Prompt caching, provisioned throughput, knowledge bases and guardrails are not counted. The calculator runs in your browser; nothing you type is sent anywhere.