Free calculator
What does AI cost your business on AWS?
Pick what you want the AI to do, how often, and which model. You get the monthly cost with the official Amazon Bedrock prices AWS publishes, refreshed every day. No sign-up.
- Official AWS prices, updated daily
- 8 models: Claude, Nova, Llama and more
- No email required
Calculate
Your monthly estimate
9,900 requests per month
Estimated model cost
US$3.96/ month
- per request
- US$0.0004
- per year
- US$48
How to pay less
- Try Amazon Nova Micro first: it would cost US$1.25 per month for the same workload, 69% less. If quality holds, keep the large model for the hard cases only.
- Turn on prompt caching: it would drop to US$2.89 per month on the same model.
- Set an AWS Budgets alert before launch: a usage spike should never show up as a surprise on the bill.
Already running AI on AWS, or about to start?
We review your account for free: permissions, costs and alerts before the bill grows.
Per-token prices from the official AWS Price List (us-east-1), retrieved on October 17, 2026. Model cost only. Vector database, storage and data transfer are not included.
Not included
What this estimate leaves out on purpose
Vector database
A document search assistant needs a vector index. Some options, like OpenSearch Serverless, bill a monthly minimum even with no traffic; others, like S3 Vectors or Aurora PostgreSQL with pgvector, scale closer to zero.
Embeddings
Turning your documents into vectors is a separate charge, usually small next to the model, paid when documents are loaded or updated.
Agents and multi-step flows
An agent may call the model several times per user question. Multiply the requests in the calculator by the steps your flow takes.
Storage, transfer and monitoring
S3 for your files, data transfer, CloudWatch logs and guardrails. Small at the start, worth budgeting for.
Method
How we calculate it
Every request sends tokens to the model (your instructions plus the question or document) and receives tokens back. Each use case has a typical size; a support chatbot, for example, uses about 2,000 input and 400 output tokens per request.
Cost = input tokens × input price + output tokens × output price, multiplied by your monthly volume. With prompt caching, the part of the input repeated on every request is billed at the cache price. With batch, the AWS batch price applies.
Prices come from the AWS Price List, the public file where AWS publishes its rates, for the US East (N. Virginia) region. Claude and Nova 2 Lite use the global inference profile; regional endpoints cost about 10% more.
| Use case | Input tokens | Output tokens |
|---|---|---|
| Support chatbot | 2,000 | 400 |
| Questions over documents | 4,000 | 500 |
| Summarize documents | 6,000 | 600 |
| Extract data | 1,000 | 150 |
Sources
FAQ
Frequently asked questions
- How much does an AI chatbot cost on AWS?
- With today's AWS prices, a chatbot handling 10,000 requests a month costs about US$1.26 with Amazon Nova Micro; US$4.00 with Claude Haiku 5.5; US$80 with Claude Sonnet 5.5 in model tokens, before prompt caching. Add the infrastructure around it: the knowledge base, storage and monitoring.
- What is a token?
- It is the unit models charge by. In English, a token is roughly three quarters of a word, so 1,000 tokens are about 750 words. You pay for the tokens you send (instructions, question, documents) and the ones the model writes back; output tokens cost several times more.
- What is Amazon Bedrock?
- It is the AWS service for using models from Anthropic, Amazon, Meta, OpenAI and others through an API, paying per use and without managing servers. Requests and data stay inside your AWS account, under your permissions and your bill.
- If each request costs so little, why do surprise AI bills happen?
- Because of volume and of what the request does not show: very long contexts, agents that call the model several times, a vector database with a monthly minimum, and no budget alerts. Most surprises come from not measuring, not from the model price.
- Where do the prices come from and how often are they updated?
- From the AWS Price List, the public file where AWS publishes its rates, for the US East (N. Virginia) region. The page reads it again twice a day. The prices you see were retrieved on October 17, 2026.
- Which model should my company use?
- Start with the smallest model that handles your real questions well and move up only where quality falls short. Test with real examples, not demos. CloudTing does not build models: we prepare the AWS account (permissions, networking, costs and alerts) so your team or your provider can run them under control.
Next step
Running AI on AWS without bill surprises
If you already use Bedrock or are about to, we review your AWS account for free: permissions, cost alerts and usage visibility per team. You get a prioritized report.
See what the free review covers →Or write to us at
hola@cloudting.com.co