AWS Builder Center
Run Your First Prompt on Amazon Bedrock in 20 Minutes (No Code, and I Show You the Bill)

Run Your First Prompt on Amazon Bedrock in 20 Minutes (No Code, and I Show You the Bill)

A hands-on, beginner-friendly walkthrough for graduating from the PartyRock playground to the real AWS service. You will send your first prompt in Amazon Bedrock, compare two foundation models on cost and quality, set a budget before you spend a cent, and pick up AI Practitioner exam concepts along the way, including tokens, model selection, and guardrails.

Improving 1% daily = 37x better per year.

What actually changes when you leave the playground

Three things change, and each one maps to something the exam cares about.
You bring your own account. No more social login. You get an AWS account, an identity, and permissions. That is IAM, and it is the reason "who is allowed to call which model" becomes a real question instead of a paragraph in a study guide.
You pay per token. Amazon Bedrock has no permanent free tier. Inference is billed from the first call, per token in and per token out. A playground abstracts this away so you can play. Bedrock exposes it so you can build.
You get the controls. Model catalog, inference parameters, guardrails, logs, cost reports. Everything a playground decided for you is now yours to decide.
Here is the good news on the money, and it is better than most people expect. New AWS accounts choose a Free plan at sign-up that starts with 100 dollars in credits, and you can earn another 100 by completing five small onboarding tasks worth 20 dollars each. One of those five tasks is testing a prompt in Amazon Bedrock. Another is setting a cost budget. In other words, this article pays for itself twice over.

What we are building

A Quiz Buddy, stripped down to its skeleton. You give it a topic, it writes five multiple choice questions, marks the correct answer, and explains in one line why that answer is right. Useful for reviewing any subject, and here we will point it straight at the content of the AI Practitioner exam, so the tool you build is the tool you study with.
No widgets, no interface. Just you, a prompt, and a foundation model, exactly as an application would call it.
That sounds like a downgrade compared to a no-code builder. It is the opposite. A no-code builder hands you a house and never lets you see the wiring. Today you see nothing but the wiring, and that is what makes the house make sense.

Step by step

1. Create your account and choose the Free plan

Go to the AWS sign-up page. When it asks whether you want the Free plan or the Paid plan, choose Free. You get the credits, you get access to the always-free services, and the plan runs for six months or until the credits run out, whichever comes first.
Yes, it asks for a card. No, choosing the Free plan does not start charging you. Have it nearby anyway.

2. Set a budget before you build anything

I am putting this at step two on purpose, before the fun part. Open AWS Budgets and create a monthly cost budget of five dollars with an email alert at eighty percent.
Five dollars is far more than this tutorial needs. The point is not the number, it is the habit. Every engineer I know who got a scary AWS bill got it because they built first and looked at the bill later. Reverse the order once and it stays reversed forever.
Bonus: this is one of the five tasks that earns you 20 dollars in credits. You just got paid to be careful.

3. Pick your Region, and understand why it matters

Top right corner of the console. If you are in Brazil, South America (São Paulo) is a legitimate choice and no longer a compromise, since it is now among the Regions where the newest Bedrock experience is available.
What to notice: not every model exists in every Region. This is the first time you will feel that AWS is a collection of physical places rather than a website. Model availability, latency, and where your data is processed all follow from this dropdown. Data residency is a Domain 5 topic on the exam, and you just made a data residency decision by clicking a menu.

4. Open Bedrock and find the playground

Search for Bedrock in the console. In the left menu, look for the model catalog and the playground.
Read this before you follow any older tutorial. You may find videos and posts telling you to visit a "Model access" page and request access to each model. That page has been retired. Serverless foundation models are now enabled automatically for your account, and access is controlled through IAM policies and service control policies instead. If a tutorial tells you to click something that no longer exists, the tutorial is old, not you.
One exception worth knowing: Anthropic models, although enabled by default, still ask you to submit a one-time usage form the first time. The playground will prompt you. Fill it in, and you are through.
You may also see an invitation to try a new Bedrock console experience, built around projects, side-by-side evaluation, and prefilled code snippets. It is excellent, and it is where you will go once you start writing code. For today, stay in the classic playground, because it puts every knob on one screen.

5. Send your first prompt

Choose a cheap, fast model to start. Amazon Nova Micro is a good first pick for exactly this reason. Open the chat playground and paste this:
Generate 5 multiple choice questions about the topic below. For each question, show 4 options (A, B, C, D), mark the correct answer, and explain in one sentence why it is correct.Topic: Generative AI fundamentals, including foundation models, tokens, and embeddings.
Run it. In a few seconds you have a quiz.
What to notice: there is no input field and no widget. The topic is just text sitting inside your prompt. In a no-code builder, that tidy little box where the user types a topic is only string concatenation wearing a nice outfit. Every AI application you will ever build does this: it takes user input, drops it into a prompt template, and sends the whole thing to a model. Writing that instruction well is what people mean by prompt engineering, and it carries real weight in Domain 2 of the exam.
If the playground offers a separate system prompt field, use it. Put the instructions there ("You are a strict certification exam writer. Never invent facts.") and leave only the topic in the user message. That separation is how real applications keep users from overwriting their own instructions.

6. Turn the knobs, and read the meter

On the right side you will find temperature, top-p, max tokens, and stop sequences. These are inference parameters, another item that shows up on the exam. Three of them are worth your time right now.
Temperature. Drop it close to 0 and run. Push it close to 1 and run again. Low makes the output predictable and precise, which is exactly what a factual quiz wants. High makes it varied and creative. You did not memorize that definition. You watched it happen.
Max tokens. Set it low, to something like 100, and run again. The model gets cut off mid-sentence. That is not a bug and not a limit on the model's intelligence. It is a ceiling you set. Tokens are the unit of everything here: what you are charged, what fits in the context window, and where the output stops.
The token counter. The playground shows you input and output tokens. Look at the number for your quiz. To make it concrete, a lightweight model like Nova Micro is priced in the range of a few cents per million input tokens and well under a dollar per million output tokens. Your quiz used maybe a few hundred. Do the division and you will find the answer is a fraction of a cent.
This is the moment most people relax. The fear of AWS costs is usually a fear of the unknown, and the number you just calculated is now known.

7. Compare two models on the same prompt

The playground lets you run more than one model side by side against the same input. Put your cheap model next to a stronger one, run the same quiz prompt, and read both outputs carefully.
You are now doing the thing the exam calls model selection. Notice the tradeoffs in the order they actually matter:
  • Quality. Are the harder questions genuinely harder, or just longer?
  • Cost. One of these models may be twenty or a hundred times the price per token.
  • Latency. Which one made you wait?
The lesson that took me too long to learn: the best model is rarely the biggest one. For generating five factual multiple choice questions, the cheap model may be entirely sufficient, and choosing it is not a compromise. It is engineering.

8. Meet Guardrails, the responsible AI domain made clickable

Read your quiz with a critical eye first. Generative models hallucinate, meaning they will state a wrong answer with complete confidence. When the output is material you plan to study from, that is not a quirk, it is a risk. Always verify what your Quiz Buddy claims before you trust it.
Bedrock gives you a real tool for part of this problem. In the left menu, open Guardrails and create one. Add a denied topic, block a category of harmful content, or turn on the filter that masks personal information such as emails and phone numbers. Then apply the guardrail in the playground and try to make your Quiz Buddy misbehave.
Watch it get blocked. That is the difference between reading "implement safeguards" in a study guide and having implemented one. Guardrails, model evaluation, and human review are the practical answers to the responsible AI domain, and you now have hands on the first of the three.

9. Go back tomorrow and look at the bill

This is the step everyone skips and the one that changes your relationship with the cloud. Tomorrow, open Cost Explorer or the billing dashboard and find your Bedrock line item.
You will almost certainly see a number so small it is funny. Look at it anyway. The engineers who are comfortable in AWS are simply the ones who have looked at the bill a hundred times.

What you just learned without memorizing anything

You did not study today. You clicked. But you now have felt, not read:
  • Prompts are instructions, and the user's input is just text dropped into them
  • Inference parameters change behavior, and max tokens truncates output
  • Tokens are the currency of both price and context
  • Foundation models are chosen, not given, and the choice is a cost and quality tradeoff
  • Regions determine model availability and where your data is processed
  • Access is governed by IAM, not by asking permission for each model
  • Guardrails are a real control you can apply in one click
  • Budgets and cost visibility are part of building, not an afterthought
Cross-reference that list with the AIF-C01 domains sometime this week. You will find you are further along than your notes suggest.


A note on accuracy: AWS moves faster than blog posts. The Bedrock console, the free tier, and model pricing all changed in the last twelve months, so verify the current details on the AWS pricing page and the Bedrock documentation before you rely on any number here, including mine. o/
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article