Amazon Web Services (AWS) is leaning into the growing trend toward custom models with a brand new service that it says will let enterprises bring more personalization and internal knowledge.
The move comes alongside the discharge of AWS's latest models as a part of its Nova family, which expands the capabilities of its reasoning models.
Nova 2 Lite, Nova 2 Pro, Nova 2 Sonic and Nova 2 Omni update the primary Nova models AWS announced last 12 months.
Nova 2 Lite is a quick, cost-effective reasoning model optimized for on a regular basis tasks that may process text, images and videos to generate text. Nova 2 Pro, which AWS said is its most intelligent reasoning model, can handle complex tasks reminiscent of coding agents, long-range planning and problem-solving. It could act as a “teacher” model for distillation projects. Nova 2 Sonic is a speech-to-speech model, while Nova 2 Omni enables organizations to generate each text and pictures from text, image and video inputs.
Nova Act, AWS's browser agent — announced as an experimental development kit in April — can also be powered by the Nova 2 models and now available to customers.
Nevertheless, it’s the custom model service, Nova Forge, that AWS is most enthusiastic about. The service gives customers the power to introduce proprietary data to a pre-trained model without fear that the model will forget its previous training.
Nova Forge allows enterprises to create custom, optimized versions of Nova models, which it calls “Novellas,” and produce them on to its Amazon Bedrock platform.
Custom model creation
Enterprises are increasingly turning to model distillation or custom models, especially with many industries selecting to create foundation models with domain-specific knowledge. But these can often be out of reach for a lot of firms, as not everyone can afford several Nvidia GPU H100s to construct models from scratch. Because of this, they turn to heavily fine-tuned open-source off-the-shelf models.
“You simply don't have an amazing approach 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 was possible? What in case you could integrate your data at the appropriate time throughout the training of a frontier model, then create a proprietary model that was only for you?”
Nova Forge employs what AWS calls “open training,” which allows developers to mix their proprietary data with an Amazon-curated dataset at every step of model development, with checkpoints during training. AWS said this implies models is not going to regress on foundational capabilities, reminiscent of instruction following, while learning company-specific knowledge and directions.
Each “Novella” might be a custom version of Nova 2 Lite, with Nova’s full knowledge and reasoning power, but with domain-specificity.
At once, enterprises can only make Novellas from Nova 2 Lite, but many will expand to other Nova 2 models soon.
Nova Forge also offers enterprises “reinforcement learning gyms." This enables them to coach AI systems through their very own environments with simulated scenarios to create smaller, faster models and access responsible AI toolkits. Once firms create their Novellas, they will bring them to Bedrock to construct more applications and agents.
One customer currently using Nova Forge is Reddit, which integrated its own data and community-specific knowledge right into a model to construct a moderation program.
Nova Forge only works with Nova models, and AWS doesn’t plan to herald third-party open-source models hosted on Bedrock (for now).
Nova 2 models intimately
AWS said tens of 1000’s of firms now use its Nova models and the corporate expects the Nova 2 models to see the identical adoption.
“Nova 2 Lite delivers incredible price performance for a lot of workloads that we actually see our customers wanting to deliver in production,” Garman said. “We predict Nova 2 Lite can be the workhorse for a lot of firms, while Pro can be for more complex tasks and for while you need your agents to be great.”
In a press release, AWS said evaluations showed Nova 2 Lite performed “equal or higher on 13 out of 15 benchmarks in comparison with Claude Haiku 4.5, equal or higher on 11 out of 17 benchmarks in comparison with GPT-5 Mini and equal or higher on 14 out of 18 benchmarks in comparison with Gemini Flash 2.5.”
Users can adjust how much Nova 2 Lite shows its step-by-step considering to balance costs with depth.
Nova Pro 2 also performed well in benchmark testing in comparison with Claude Sonnet 4.5, GPT-5.1 and Gemini 2.5 Pro. This model works best for multi-document evaluation, video reasoning, advanced math and agentic engineering tasks.
AWS said in its press release that each Nova 2 Lite and Pro “have built-in grounding and code execution capabilities.”
Nova 2 Sonic, the speech-to-speech model, generates human-like conversations and now supports multiple languages. The updated model has a 1-million-token context window, with more expressive voices and better accuracy. The corporate said Sonic may even switch topics mid-conversation.
Nova 2 Omni handles “as much as 750,000 words, hours of audio, long videos and hundred-page documents, concurrently analyzing entire product catalogs, testimonials, brand guidelines and video libraries without delay.”
“While there are not any comparable models within the industry to Nova 2 Omni, it demonstrates strengths in public benchmarks of multimodal reasoning on documents, images, videos and audio, and may generate high-quality images much like other leading image-generation models,” AWS said in its release.
