AWS Bedrock Cost Calculator

An AWS Bedrock cost calculator is a tool that estimates the on-demand inference expenses of running foundation model API calls through Amazon Bedrock. AWS Bedrock charges per token — both for the input (your prompt) and output (the model's response) — at rates that vary by model provider and model tier. Our AWS Bedrock cost calculator online processes these variables locally in your browser, letting developers, architects, and finance teams project API spend before deploying workloads to production without sending any data to a server.

1. Select a Bedrock Model: Choose from a curated list of supported foundation models grouped by provider — Anthropic Claude, Amazon Nova, Meta Llama, Mistral AI, and Cohere. The calculator pre-loads the current US-East-1 on-demand token rates for each model automatically.
2. Enter Token Counts: Specify your average input tokens per request (system prompt + context + user message) and expected output tokens (model response length). Use the interactive sliders for quick rough estimates or type exact values.
3. Configure Cost Optimizations: Enable Prompt Caching to reduce costs on repeated static context (like large system instructions or RAG documents), or toggle Batch Inference to apply the 50% discount for asynchronous background processing workloads.
4. Review Forecasts and Comparisons: Check the projected cost per single call, along with daily, monthly, or yearly totals based on your request volume. Use the provider-filtered comparison table to identify the cheapest Bedrock model that fits your context window requirements.

• On-Demand vs. Provisioned Throughput: On-demand rates (used in this calculator) charge per token with no commitment. Provisioned Throughput Units (PTUs) reserve dedicated capacity for a fixed hourly fee — typically 40–60% cheaper for sustained high-volume workloads above ~200M tokens/month.
• Input vs. Output Token Costs: Output tokens are generated token-by-token and are computationally more expensive. Most Bedrock models price output tokens 3×–10× higher than input tokens. Minimizing unnecessary output length is the highest-leverage cost reduction lever.
• Prompt Caching Discount: AWS Bedrock supports prompt caching for select models. Cached tokens (e.g., a long static system prompt or reference document) are billed at a steep discount — up to 90% less than standard input rates — making it highly effective for repeated-context workloads like chatbots and agents.
• Batch Inference Pricing: Bedrock Batch Inference processes large collections of requests asynchronously (results within 24 hours) at exactly 50% of standard on-demand prices. Ideal for bulk document processing, dataset evaluation, and overnight summarization pipelines.

To reduce AWS Bedrock costs: first, test your use case on lighter models like Amazon Nova Micro, Claude 3.5 Haiku, or Mistral Small before assuming a premium model is needed — classification, routing, and simple extraction tasks often perform equally well at 10–50× lower cost. Second, enable prompt caching if your system prompt or RAG context is longer than 1,024 tokens and repeated across many requests. Third, route any non-latency-sensitive pipeline (batch translation, offline indexing, nightly evaluations) through Bedrock Batch Inference for an automatic 50% discount. Finally, include Bedrock token spend in your AWS EDP commitment calculations, as it counts toward your enterprise discount volume thresholds.