AWS Launches Nova Forge to Power Custom Training of Cutting-Edge "Novella" Models

Amazon Web Services (AWS) today unveiled Nova Forge at the AWS re:Invent conference—a groundbreaking new service enabling enterprises to train and build their own state-of-the-art AI models based on AWS’s Nova foundation model series.

The cloud computing leader has not yet disclosed official pricing but indicated to CNBC that the service will cost approximately $100,000 annually.

Nova is a family of multimodal foundation models developed and released by AWS. The lineup includes the newly launched Nova 2 Lite, Nova 2 Pro, and Nova 2 Omni—each introduced alongside Nova Forge.

As businesses increasingly fine-tune or extend existing models with proprietary data, they face a growing challenge: the more domain-specific data a model absorbs, the greater the risk it “forgets” aspects of its original training. This phenomenon, known as catastrophic forgetting, can degrade core capabilities such as instruction following. According to AWS, organizations need a way to create cutting-edge custom models without compromising these foundational skills.

AWS emphasizes that enterprises require a reliable method to develop advanced AI models while avoiding performance degradation.

Nova Forge addresses this need. Custom models built through the service—dubbed “Novellas”—fuse enterprise-specific data with the advanced capabilities of the Nova series. The platform grants exclusive access to pre-training, mid-training, and post-training checkpoints of Nova models, allowing customers to blend their data with Amazon’s curated datasets for optimal training outcomes.

Enterprises can begin building Novellas using Nova 2 Lite—the smallest model in the series, optimized for fast and cost-efficient inference on everyday tasks. Nova Forge users also gain early access to Nova 2 Pro and Nova 2 Omni, giving them a head start in developing applications and more powerful Novellas ahead of general availability.

Beyond data integration, Nova Forge introduces several advanced capabilities. Customers can train AI models within their existing environments—referred to as reinforcement learning “gyms”—where models interact with synthetic data and simulated scenarios that mirror real-world usage.

Developers can also produce distilled models that are smaller and faster while preserving much of the intelligence of larger counterparts, all at reduced cost. Additionally, AWS bundles an AI Safety Toolkit, empowering businesses to implement safeguards and policy-aligned controls.

Early adopters of Nova Forge include Booking.com B.V., Cosine AI, Nimbus Therapeutics, Nomura Research Institute, OpenBabylon, Sony Group Corporation, and Reddit Inc.—all leveraging the service to build tailored AI solutions aligned with their unique operational needs.

“We’re replacing multiple disparate models with a single, more accurate solution, making content moderation significantly more efficient,” said Chris Slowe, CTO of Reddit. “Moving from several specialized machine learning workflows to one cohesive approach marks a pivotal shift in how we deploy and scale AI across Reddit.”

Nova Forge is now available in the AWS US East region, with plans to expand to additional regions in the coming months.