Amazon Web Services (AWS) is leaning into the growing trend toward custom models with a new service that it says will let businesses bring more personalization and internal intelligence.
The move comes alongside the release of new AWS models as part of its Nova family, which expands the capabilities of its inference models.
The Nova 2 Lite, Nova 2 Pro, Nova 2 Sonic and Nova 2 Omni are the first updated Nova models. It was announced last year.
The Nova 2 Lite is a fast, cost-effective reasoning model optimized for everyday tasks that can process text, images and videos to generate text. Nova 2 Pro, which AWS said is its most intelligent reasoning model, can handle complex tasks such as coding agents, long-range planning and problem solving. It can serve as a “teacher” model for Asan projects. The Nova 2 Sonic is a speech-to-speech model, while the Nova 2 Omni enables organizations to generate both text and images from text, image and video input.
Nova Act, AWS’s browser agent – announced as Experimental Development Kit in April – is also powered by Nova 2 models and is now available to users.
However, it’s the custom model service, Nova Forge, that AWS is most excited about. The service gives users the ability to introduce proprietary data into a pre-trained model without fear that the model will forget its previous training.
Nova Forge allows businesses to create custom, customized versions of Nova models, which they call “novels,” and bring them directly to their Amazon Bedrock platform.
Custom model creation
Enterprises are turning fast Model flexibility or Custom modelespecially with many industries choosing to build a foundation model Domain related knowledge. But these can often be out of reach for many companies, as not everyone can afford to model the Nvidia GPUH100S from scratch. As a result, they turn away from the off-the-shelf model to heavily leveraged open source.
“You don’t have a good way to get a frontier model that deeply understands your data and your domain," AWS CEO Matt Garman said during his keynote speech at AWS’s annual RE:Invent conference. "But what if it were possible? What if you could integrate your data at the right time while training a frontier model, then create a proprietary model that was just for you?
Nova Forge calls AWS “open training,” which allows developers to blend their proprietary data with Amazon-sourced datasets at each stage of model development, with checkpoints during training. AWS said this means models will not learn basic skills, such as following directions, while learning company-specific information and instructions.
Each “novel” can be a custom version of Nova2 Lite, with the full information and reasoning power of Nova, but with domain specificity.
Right now, businesses can only make novelties with the Nova 2 Lite, but many will soon expand to other Nova 2 models.
Nova Forge Enterprises also offers “assistive learning gyms.”" It allows them to train AI systems through simulated scenarios through their environment to build smaller, faster models and access responsive AI toolkits. Once companies create their novel compounds, they can bring them to bedrock to build more applications and agents.
One user currently using NovaForge is Reddit, which integrated its data and community-specific knowledge into a model to build a moderation program.
Nova Forge only works with Nova models, and does not plan to bring third-party open source models hosted on AWS Bedrock (for now).
Nova 2 model in detail
AWS said tens of thousands of companies now use its Nova models, and the company expects the Nova 2 models to see similar adoption.
“Nova 2 Lite delivers incredible price performance for many of the workloads we see our customers looking to deliver in production,” said Garmin. “We think the Nova 2 Lite will be the workhorse for many companies, while the Pro will be for more complex tasks and when you need your agents to be great.”
In a press release, AWS said that evaluations showed that the Nova 2 Lite performed equal or better on 13 of 15 benchmarks against Cloud HiCo 4.5, equal or better on 11 of 17 benchmarks against GPT5 Mini, and equal or better on 14 of 18 benchmarks against Gemini Flash 2.5. “
Users can adjust how deeply the Nova 2 Lite displays its step-by-step thinking to balance costs.
The Nova Pro 2 also performed well in benchmark testing against the CloudSonata 4.5, GPT-5.1 and Gemini 2.5 Pro. This model works best for multi-document analysis, video reasoning, advanced mathematics and agentic engineering tasks.
Both Nova 2 Lite and Pro “have built-in grounding and code execution capabilities,” AWS said in its press release.
Nova 2 Sonic, the speech-to-speech model, produces human-like conversations and now supports multiple languages. The latest model has a 1 million token context window, with more expressive sounds and higher accuracy. Sonic can even change topics mid-exchange, the company said.
Nova 2 Omni “analyzes entire product catalogs, testimonials, brand guidelines and video libraries simultaneously, up to 750,000 words, hours of audio, long videos and hundreds of pages of documents.”
“Although there are no comparable models in the industry to the Nova 2 Omni, it demonstrates strength in public standards for multimodal reasoning on documents, images, videos and audio, and can produce high-quality images similar to other leading image generation models,” AWS said in its release.