Building intelligent agents to handle complex, real-world tasks can be difficult. Additionally, instead of relying solely on large, pre-trained base models, organizations often need to tune and customize smaller, specialized models to work best for their specific use cases. The AWS AI League provides an innovative program to help businesses overcome the challenges of building advanced AI capabilities through exciting competitions that drive agent AI innovation and model customization.
In 2025, the first AWS AI League competition attracted the attention of engineers, data scientists, and business leaders around the world. They come together to solve pressing problems using the latest AI tools and techniques. The grand finale at AWS re:Invent 2025 was an exciting showcase of their creativity and talent. Collaborative teams from leading organizations compete head-to-head, demonstrating their ability to execute effective commands, fine-tune models, and build powerful AI agents.
Congratulations to our 2025 AWS AI League Champions! After intense competition between these three outstanding builders, they won, sharing the $25,000 prize pool:
- First place: Hemanth Vediyera from Cisco
- Second place: Ross Williams of Aqfer
- 3rd Place: Deepesh Khanna from Capital One
Figure 1: Left to right: Ross, Hemanth, Deepesh
This post explores how the AWS AI League system can be used to host AI competitions that can help participants discover customization model and agent building concepts, apply them to real-world business challenges, and demonstrate their innovative solutions in engaging, game-style formats. We highlight new agent AI and customizable modeling challenges, where businesses can apply to host internal competitions using AWS credits, and developers can compete in AWS events.
To get started, visit the AWS AI League product page.
What is the AWS AI League Championship?
The AWS AI League experience begins with a 2-hour hands-on workshop led by AWS experts, followed by a self-assessment. The journey culminates in a grand, game show-style finale, where you showcase your AI creations and solutions to pressing business challenges. The following figure shows these three steps.
Figure 2: Steps to the AWS AI League Championship
Building on the success of the 2025 edition, we are excited to announce the launch of the AWS AI League 2026 Championship. This year, the competition features two new challenges that allow participants to test their AI skills:
- The agent AI Challenge allows you to build intelligent agents using Amazon Bedrock AgentCore. Competitors are building customized architectural agents to address real-world business problems.
- In line with the agency's AI Challenge, the custom modeling challenge now uses the latest fine-tuning recipes in SageMaker Studio. Here you customize models for specific use cases.
For the 2026 AI League, the prize money doubles to $50,000, with tracks catering to developers of different skill levels – from beginners to advanced practitioners.
Build smart agents with the agent AI challenge
The AWS AI League now has an exciting AI agent challenge, where you build intelligent agents using Amazon Bedrock AgentCore to solve complex problems in a dynamic, game-style competition. In this challenge, agents navigate a grid maze-like environment, encountering various challenges while searching for a treasure chest. These challenges target real-world use cases, testing an agent's ability to handle negative content, code execution, browser usage, and more.
Agents have a time limit to traverse the map, collect points, and overcome obstacles before reaching the treasure chest. The more points they earn, the higher they go on the leaderboard. You can fully customize your agents using Amazon Bedrock AgentCore primitives, which enable you to safely scale and manage production-grade agents. You can also select specific manager and sub-agent models, and create custom tools like Bedrock Guardrails, AgentCore Memory, and AWS Lambda functions to help your agents navigate challenges. The following figure shows the obstacles the agent must overcome while traveling to reach the treasure chest.
Figure 3: AWS AI League Agetic Challenge
AWS AI League provides a full user interface (UI) for users to build their own intelligent agent solutions. You can use this code-free UI to build multi-agent architectures and tools, integrating various components such as Amazon SageMaker Studio CodeEditor to code collaboratively for custom Lambda functions and tools. This allows you to fully develop and customize your agent-based solutions within the AWS AI League website, without the need to leave the environment.
The following screenshots show the experience of building an agent all within the AWS AI League website.
Figure 4: AWS AI League agent tools
Figure 5: AWS AI League multi-agent architecture
Throughout the competition, users get real-time agent performance feedback, with a large-scale language model (LLM) tester that provides testing to help with iteration. The following image shows how the agent is evaluated during challenges.
Figure 6: Evaluation of the AWS AI League agent challenge
In the grand competition, the top finalists take the stage to showcase their agents' abilities in a live, game show format, demonstrating the power and flexibility of agent AI in solving complex, multi-step problems. Evaluation criteria include time efficiency, accuracy in solving challenges, scheduling agents, and token efficiency. The following summary shows the last round of the Grand Finale at re:Invent 2025.
Figure 7: AWS AI League re:Invent 2025 Grand Final
Customize models to perform better than large models
The AWS AI League expands the scope of its model customization challenge, allowing you to use the latest advances in fine-tuning techniques.
You can access the new customization model within Amazon SageMaker Studio, where you can use powerful new training recipes. The goal is to develop highly efficient, domain-specific models that can outperform larger, reference models.
The challenge starts with honing your modeling skills. Using the tools and techniques you've learned, you use advanced fine-tuning techniques to help improve your model's performance. After your models are customized, the real testing begins. Models are submitted to a leaderboard to evaluate performance against a reference model. The model earns points each time the automated judge deems your customized model's answer to be more accurate and comprehensive than the reference model's output. You can show off your advanced skills, climb to the top of the leaderboard, and open up new opportunities for your organizations.
During the challenge, you get real-time feedback on your model's performance from an automated tester when you post to the leaderboard. The leaderboard evaluates submissions against a reference dataset across the competition, providing quick and accurate feedback to help you iterate and improve your solutions. The following image shows how AI critique is used to test a customized model.
Figure 8: AWS AI League model customization test
In the main competition, the top finalists show off their modeling skills in a live, game show format, showcasing their rapid engineering skills. During the game show, scores include expert evaluations where domain experts and a live audience participate in real-time voting to determine which AI solutions best solve real business challenges. The following image shows the view of the participating engineers during the Grand Finale.
Figure 9: AWS AI League model for customizing a participant's view of the Grand Finale
The conclusion
In this post, we explore the new challenges of the AWS AI League and how they are changing the way organizations approach AI development. At AWS, we've learned that the fastest way to spark innovation is through competition. With AWS AI League, developers can now show their AI skills, compete and unlock new things.
To learn more about hosting AWS AI League at your organization visit AWS AI League and dive deeper into building intelligent agents and customizing AI models check out the AWS AI training catalog in the AWS Skill Builder.
About the writers
Marc Karp he is an ML Architect with the Amazon SageMaker Service team. He specializes in helping customers design, deploy, and manage ML workloads at scale. In her spare time, she likes to travel and explore new places.
Natasya K. Idries He is the Product Marketing Manager for AWS AI/ML Advanced Learning Systems. He is passionate about democratizing AI/ML skills through engaging and collaborative educational programs that bridge the gap between advanced technology and practical business implementation. His expertise in building learning communities and driving digital innovation continues to shape his approach to creating impactful AI education programs. Outside of work, Natasya likes to travel, cook Southeast Asian food and explore nature trails.
