Amazon unveils Nova 2 AI models & launches custom Forge tools

Amazon has launched new additions to its Nova AI portfolio, introducing four Nova 2 models designed for advanced reasoning, multimodal understanding, and generative tasks across text, images, video, and speech. The company also unveiled Nova Forge, a service allowing organisations to build their own bespoke AI models, as well as Nova Act for developing reliable AI agents for browser-based automation.

Model portfolio

The new Nova 2 model family includes Lite, Pro, Sonic, and Omni variants. Each aims at specific use cases across industries such as customer service, document processing, code generation, and interactive voice experiences. Nova 2 Lite provides fast, cost-effective reasoning for day-to-day workloads and supports adjustment of reasoning depth, making it suitable for tasks like chatbots and business automation. Nova 2 Pro is targeted at more complex reasoning, handling tasks like agentic coding, long-range planning, and knowledge distillation into smaller models for specific domains.

Both Lite and Pro versions are built with web grounding and code execution capabilities, enabling access to current information and the ability to run code as part of their responses. Nova 2 Sonic serves real-time conversational applications with advanced multilingual speech support, while Nova 2 Omni offers unified multimodal generation, able to handle large-scale text, image, video, and audio analysis and create outputs in both text and image formats.

Industry applications

Major organisations including Cisco, Siemens, Sumo Logic, and Trellix are already using Nova 2 models for automated content creation, video analysis, threat detection, and intelligent voice assistants. The models demonstrate strengths in public benchmarks for tasks such as document processing, code generation, multi-step workflow automation, and multimodal reasoning.

Custom model development

Nova Forge enables companies to create their own tailored versions of Nova-referred to as "Novellas"-by integrating proprietary datasets during model training. This "open training" approach lets customers inject exclusive data at various stages and provides access to model checkpoints, allowing fine-tuning and specialised optimisation. The service also supports reinforcement learning in synthetic environments and includes responsible AI tools for safety and governance.

Organisations such as Booking.com, Cosine AI, Nimbus Therapeutics, Nomura Research Institute, OpenBabylon, Reddit, and Sony are among those building custom models with Nova Forge. They gain the ability to mix their proprietary knowledge with large-scale AI capabilities and deploy these custom models securely on Amazon Bedrock, benefiting from established AWS security, scalability, and data privacy practices.

"Working with Nova Forge is allowing us to improve content moderation on Reddit with a more unified system that's already delivering impressive results. We're replacing a number of different models with a single, more accurate solution that makes moderation more efficient. The ability to replace multiple specialised ML workflows with one cohesive approach marks a shift in how we implement and scale AI across Reddit. After seeing these early successes in our safety efforts, we're eager to explore how Nova Forge might help in other areas of our business," said Chris Slowe, CTO, Reddit.

Agent automation

Nova Act, now an AWS service, targets the creation and deployment of AI agents able to automate tasks within web browsers. Running on a custom-trained Nova 2 Lite model and using reinforcement learning, Nova Act has been benchmarked at 90% reliability for early customer workflows including data updates, website functionality testing, and automated claims processing. Developers can prototype and scale agents quickly, using no-code tools and familiar development environments such as VS Code.

Early adopters include startups and enterprises such as Sola Systems, 1Password, and Hertz. Applications range from workflow automation to end-to-end software testing and digital assistant deployment. Organisations report significant reductions in manual workload, improved speed, and greater reliability in business-critical processes.