💸 What Does One Amazon Bedrock Prompt Cost? Find Out From the CLI (Hands-on)

#machinelearning #ai #aws #tutorial

🎤 The interview question

"Your team wants to add an Amazon Bedrock chatbot. How would you work out what one prompt costs?"

It shows up in AIF-C01 prep and in real interviews. The strong answer: Bedrock charges per token, and every response tells you how many tokens it used. Let's prove that from the AWS CLI in 5 minutes. 👇

👉 Flow: Send one prompt → Read usage → Multiply by the price → Compare two prompts

🔧 Step 1: Set up

✅ AWS CLI v2, signed in (aws sts get-caller-identity works)
✅ Region us-east-1
✅ Model access for Amazon Nova Micro (Bedrock console → Model access)

💬 Step 2: Send one prompt with the Converse API

aws bedrock-runtime converse \
  --region us-east-1 \
  --model-id us.amazon.nova-micro-v1:0 \
  --messages '[{"role":"user","content":[{"text":"Explain overfitting in one sentence."}]}]' \
  --query '{answer: output.message.content[0].text, usage: usage}'

✅ Expected (your numbers will differ):

{
  "answer": "Overfitting is when a model learns the training data so closely ...",
  "usage": { "inputTokens": 8, "outputTokens": 24, "totalTokens": 32 }
}

👀 That usage block is your bill, in tokens.

🧮 Step 3: Turn tokens into money

Open Amazon Bedrock pricing, find Nova Micro, and copy its price per 1,000 input tokens and per 1,000 output tokens.

IN=8; OUT=24                      # from usage
P_IN=<price per 1K input tokens>  # from the pricing page
P_OUT=<price per 1K output tokens>
awk -v i=$IN -v o=$OUT -v pi=$P_IN -v po=$P_OUT \
  'BEGIN { printf "USD %.8f per prompt\n", i/1000*pi + o/1000*po }'

💡 Multiply by prompts per day to get a daily cost. That's the number your manager wants.

🧨 Break it on purpose

Ask for a long answer instead:

--messages '[{"role":"user","content":[{"text":"Explain overfitting in 500 words."}]}]'

📈 outputTokens jumps, and output tokens cost more than input tokens. Then add --inference-config '{"maxTokens":50}' and watch the cost come back down.

🧹 Cost and cleanup

💵 A few prompts on Nova Micro cost a fraction of a cent
🧹 Nothing to delete: Converse calls create no resources

🎯 What you can now say in the interview (and the exam)

✅ "Bedrock on-demand pricing is per input and output token"
✅ "Every Converse response returns inputTokens and outputTokens"
✅ "Output tokens usually cost more, so I cap them with maxTokens"
✅ "Cost per prompt × prompts per day = the daily bill"

📚 Go deeper

This is the hands-on cut of Chapter 5 of our free AWS AI Practitioner (AIF-C01) course: the full chapter explains tokens, why they drive cost, and how the exam asks about them.

👉 Read the full chapter on Confident Prep

DEV Community