Amazon Online Services (AWS) is part of the growing trend toward custom models with a novel service it says will allow enterprises to provide greater personalization and in-house knowledge.
This move comes with the release of novel AWS models as part of the Nova family, which expands the capabilities of reasoning models.
Nova 2 Lite, Nova 2 Pro, Nova 2 Sonic and Nova 2 Omni update the first Nova AWS models announced last year.
Nova 2 Lite is a speedy and cost-effective reasoning model optimized for everyday tasks that can process text, images and videos to generate text. Nova 2 Pro, recognized by AWS as the smartest reasoning model, can handle convoluted tasks such as coding agents, long-range planning, and problem solving. It can act as a “teacher” model in distillation projects. Nova 2 Sonic is a speech-to-speech model, while Nova 2 Omni enables organizations to generate both text and images from text, image and video inputs.
Nova Act, AWS Browser Agent – Announced as experimental development kit in April — is also powered by Nova 2 models and is now available to customers.
But it’s the custom model service, Nova Forge, that AWS is most excited about. The service gives customers the opportunity to enter their own data into a previously trained model without fear that the model will forget the previous training.
Nova Forge enables enterprises to create custom, optimized versions of Nova models, which it calls “Novellas,” and bring them directly to the Amazon Bedrock platform.
Creating custom models
Enterprises are increasingly turning to model distillation Or custom modelsespecially in the case of many industries choosing to create foundation models domain-specific knowledge. However, these can often be out of reach for many companies, as not everyone can afford multiple Nvidia H100 GPUs to build models from scratch. As a result, they are turning to highly tuned, off-the-shelf, open-source models.
“There is simply no good way to get a boundary model that deeply understands your data and your domain,” AWS CEO Matt Garman said during his keynote address at the annual AWS re:Invent conference. “But what if it were possible? What if you could integrate your data at the right moment while training the pioneer model, and then create your own model that was just for you?”
Nova Forge uses what AWS calls “open training,” which allows developers to combine their proprietary data with a dataset selected by Amazon at any stage of model development, with checkpoints during training. AWS says this means models won’t regress in basic capabilities, such as executing instructions, while learning company-specific knowledge and instructions.
Each “Novella” can be a custom version of Nova 2 Lite, with all of Nova’s knowledge and reasoning power, but with domain specificity.
At the moment, companies can only produce Novell models with the Nova 2 Lite, but many will soon expand their offerings to other Nova 2 models.
Nova Forge also offers “empowerment training rooms” to businesses. This allows them to train AI systems in their own environments using simulated scenarios to create smaller, faster models and access responsible AI toolkits. Once companies create their novels, they can move them to Bedrock to build more applications and agents.
One of the clients currently using Nova Forge is Reddit, which has integrated its own data and community-specific knowledge into the model to build a moderation program.
Nova Forge only works with Nova models, and AWS has no plans to introduce third-party open source models hosted on Bedrock (yet).
Details of Nova 2 models
AWS says tens of thousands of companies currently operate Nova models, and the company expects the Nova 2 models to see the same reception.
“Nova 2 Lite delivers incredible price performance for many of the workloads our customers want to run in production,” Garman said. “We think Nova 2 Lite will be a workhorse for many companies, while Pro will be intended for more complex tasks and when you need your agents to be great.”
In a press release, AWS said the evaluations showed the Nova 2 Lite performed “equal or better in 13 of 15 benchmarks compared to Claude Haiku 4.5, equal or better in 11 of 17 benchmarks compared to GPT-5 Mini, and equal or better in 14 of 18 benchmarks compared to Gemini Flash 2.5.”
Users can adjust the degree to which Nova 2 Lite shows its step-by-step thinking to balance cost with depth.
The Nova Pro 2 also performed well in benchmarks against Claude Sonnet 4.5, GPT-5.1 and Gemini 2.5 Pro. This model works best for multi-document analysis, video inference, advanced math tasks, and agent-based engineering.
AWS said in its press release that both the Nova 2 Lite and Pro “have built-in grounding and code execution capabilities.”
Nova 2 Sonic, a speech-to-speech model, generates human-like conversations and now supports multiple languages. The updated model has a context window of 1 million tokens, with more expressive votes and greater accuracy. The company claims that Sonic can even change the topic mid-conversation.
Nova 2 Omni handles “up to 750,000 words, hours of audio, long-form videos, and 100-page documents while analyzing entire product catalogs, testimonials, brand guidelines, and video libraries.”
“While there are no models in the industry comparable to Nova 2 Omni, it demonstrates strengths in public benchmarks for multimodal inference across documents, images, video and audio, and can generate high-quality images similar to other leading image generation models,” AWS said in its announcement.