Amazon Nova AI Challenge returns for a second season with Nova Forge access

The Amazon Nova AI Challenge returns for its second year, with 10 university teams from around the world gathering at a Seattle bootcamp on February 2-4 to build AI agents that can handle complex coding tasks while maintaining security and reliability.

This year's competition centers not on single code generation, but on multi-step software development tasks spanning planning-implementation-verification. Teams are evaluated on two axes: utility, which looks at whether the agent can handle harder development challenges, and safety, which looks at whether it maintains appropriate safeguards.

The biggest change is the introduction of Nova Forge. University teams get access to model checkpoints, training resources, and tools, allowing them to feed data into the training process itself and customize models, and Amazon calls these customized models "Novella."

Forge provides three capabilities.

• Custom training environments: Building simulation environments that reflect real secure coding workflows
• Model compression: Learning from examples of larger models to compress into smaller, faster models
• Safety controls: Applying security measures and evaluating against safety standards

Participating teams were selected from 5 countries: the United States, Portugal, Czech Republic, Korea, and Taiwan. Model development teams include PurpCorn (University of Illinois Urbana-Champaign), BlueTWIZ, AlquistCoder, BruinWeb, and Slugs and Roses, while red teams include PurCL (Purdue University), Jay'lBreak, TeamSecLab, Pr1smCode, and Lion-x0a.

Each team receives $250,000 in sponsorship funding and monthly AWS credits, and the winning model development team and red team will each receive $250,000, while runner-up teams receive $100,000. The first evaluation round begins after the bootcamp, with additional rounds continuing through August 2026, the finals in September 2026, and award announcements in October 2026.