If you're gearing up for the AWS 10,000 AIdeas Competition (running December 5, 2025 to January 21, 2026), you're probably thinking about two things: building something awesome and not breaking the bank while you experiment.
Well I have some good news for you builders: Alongside the new AWS Free Tier, if you want to inject large language models (LLMs) into your application and take advatange of state of the art frontier models from folks like Anthropic, Meta, Amazon and others, Amazon Bedrock will be your best friend. With the new AWS Free Tier you've got some runway to build without worrying about surprise bills. Let's walk through how to maximize your free credits to build a competitive application without emptying your wallet.
What's This Hackathon All About?
The Global 10,000 AIdeas Competition is one of the largest developer competitions in AWS history, inviting developers worldwide to build innovative AI applications using AWS tools. You can join the competition to compete for $250,000 in cash prizes as well as recognition at AWS re:Invent 2026 and across AWS channels.
The requirement is simple: pitch an application idea for one of the five competition tracks: workplace efficiency, daily life enhancement, commercial solutions, social impact, or creative expression. Your initial submission is a written pitch—no code required at this stage—that describes what you want to build and why it matters.
If selected as a semi-finalist, that's when you'll build your app. Your finished app needs to use the agentic IDE Kiro for at least part of development, stay within AWS Free Tier limits, and be completely original and not yet published. For more details and FAQs about the competition you can head to the announcement post.
The New Free Tier: Everything You Need to Know
Now let's dig into how to use free tier for this competition. AWS recently revamped Free Tier for new accounts (created after July 15, 2025), so here's how it breaks down:
- $100 in credits when you sign up
- Up to $100 more in credits by completing five onboarding activities (including one specifically for Bedrock!)
- Total: Up to $200 in credits
- Duration: 6 months (or until you exhaust your credits)
One of those extra $20 credit activities? Using Amazon Bedrock in the playground. So literally just experimenting with LLMs in the Bedrock console earns you credits. How easy is that?!
It's important to note that Amazon Bedrock itself is not in the Always Free tier—you will be using your free credits to pay for Bedrock usage if you chose to add this service to your application.
How Amazon Bedrock Pricing Works
Before we dive into optimization, let's quickly break down how Bedrock charges you (because understanding this is half the battle):
- Text models: Pay per token (both input and output)
- Image models: Pay per image generated
- Embeddings: Pay per input token
Bedrock offers three main pricing tiers:
- On-Demand - Pay as you go, no commitments (this is what you'll use)
- Batch Mode - 50% discount for non-urgent workloads using certain models
- Provisioned Throughput - Reserved capacity (this is overkill for this competition)
For this competition, we recommended you use only on-demand standard tier pricing, which means every API call counts. The good news? You can predict costs pretty accurately since these calculations are token-based.
If you're curious, you can check out the official pricing page for Amazon Bedrock which includes even more details than the ones I mention here.
Which Models Should You Use?
Here's where some strategy comes in. Not all models are created equal when it comes to cost, and choosing wisely can make your $200 credits go way further.
Budget-Friendly Champions 🏆
All prices in this blog are referencing the US-East-1 (N. Virginia) region.
Amazon Nova Micro ($0.000035/1K input tokens, $0.00014/1K output tokens)
- The cheapest of the Amazon models
- Great for simple tasks, routing, or classification
Amazon Nova Lite ($0.00006/1K input tokens, $0.00024/1K output tokens)
- Still incredibly affordable
- Solid reasoning capabilities
- My pick for most hackathon use cases where you need good performance without burning through credits
Meta Llama 3.3 70B ($0.00072/1K input tokens, $0.00072/1K output tokens)
- Great price-to-performance ratio
- Strong at reasoning and tool use
- Flat pricing (same rate for input and output) makes cost prediction easy
Where to Splurge (Strategically) 🤑
Claude Sonnet 4.5 ($0.0033/1K input tokens, $0.0165/1K output tokens)
- More expensive, but incredibly capable
- Use for complex reasoning tasks where quality really matters
- Great for your demo/presentation application where you want to wow the judges
Claude Haiku 4.5 ($0.0011/1K input tokens, $0.0055/1K output tokens)
- Faster and cheaper than Sonnet
- Good middle ground for production-like scenarios
- My pick for a model if you're wanting to splurge
I'll leave a knowledge nugget for those wanting more info on Claude Haiku 4.5, this deep dive article by Anthropic shows just how performant and affordable this model is.
The Power Move: Intelligent Prompt Routing
Here's a pro tip that can save you up to 30% on costs: use Bedrock's Intelligent Prompt Routing. This feature automatically routes simple queries to cheaper models (like Nova Lite or Haiku) and complex queries to more capable models (like Sonnet or Nova Pro).
For a customer service agent, this means basic questions like "What are your hours?" go to the cheap model, while "I need help troubleshooting this complex integration issue" gets routed to the smart (expensive) model.
There is a cost associated with this cool feature, but at $1 per 1,000 requests it might be worth experimenting with it! For more detailed info check out the documentation page.
Maximizing Your Free Credits: Battle-Tested Tips
1. Start with Nova Lite or Nova Micro
Build your application with these models first. Get the logic working, nail your tool integrations, and perfect your prompts. Only upgrade to more expensive models when you're confident your application works well.
2. Use Batch Mode for Testing
If you have large test suites or want to evaluate multiple prompt variations, try using Batch mode for a 50% discount compared to on-demand. It's not real-time, but for development and testing, it can be quite useful.
3. Optimize Your Prompts
Shorter prompts = fewer tokens = lower costs. Every word counts. Be concise in your system prompts and examples.
4. Implement Prompt Caching
Avoid making redundant API calls. If your agent asks the same question multiple times, cache the response. This feature might not be available with all models so double check the documentation to learn more.
5. Monitor Costs in Real-Time
Head to the AWS Cost Explorer and filter by:
- Granularity: Daily
- Dimension: Usage Type (This shows exactly which models are costing you money)
- Service: Bedrock
Check this daily during development. I can't stress enough how easy this is to do, you don't want to burn through $150 of credits on day three and realize it on day ten...
6. Set Up Budget Alerts
Go to AWS Budgets and create a budget.
For this specific competition you can set something up to simulate the burn rate of your free tier credits.
Create a budget for $200 with alerts at:
- 50% ($100 spent)
- 75% ($150 spent)
- 90% ($180 spent)
This way you get early warning before things get out of hand.
⚡️Bonus: setting up a cost budget is another one of those five onboarding tasks for the new Free Tier, so you automatically unlock another $20/$100 in extra AWS Credits by doing so!
7. Use the Playground First
The Bedrock playground in the console lets you test prompts without writing code. It's the same cost, but way faster for iteration. Plus, remember—using it earns you $20 in credits!
Regional Considerations
Not all models are available in all regions. The most model variety is in:
- us-east-1 (N. Virginia) - Best option for maximum model access
- us-west-2 (Oregon) - Good alternative with most models
Stick to these regions for the competition if you can to avoid any "model not available" headaches.
Final Thoughts
I know this may seem like a lot, so if you're more of a visual/video learner we actually have FREE, live learning sessions this week from AWS experts that will cover how to use AWS Free Tier!
Click below to register and attend one that fits your schedule:
Americas - January 6, 12 pm PT (GMT-8) Register
Europe, Middle East, Africa - January 7, 12 pm London Time (GMT+0) Register
Asia Pacific, Japan, China - January 7, 12 pm AEDT (GMT+11) Register
Now go build something cool. And seriously, set up those budget alerts first. Your wallet will thank you 😉
Additional Resources 📚
Top comments (0)