Amazon announces a comprehensive expansion of its Nova portfolio with four new models, a pioneering “open training” service that empowers organizations to build their custom model variants with Nova, and a service for creating highly reliable AI agents.

Tens of thousands of companies are using Nova for diverse applications, such as producing high-quality content, automating multi-step tasks, and accelerating development of AI agents. Amazon’s new Nova 2 model family balances speed, cost, and intelligence across multiple AI domains:

Nova 2 Lite is a fast, cost-effective reasoning model for everyday workloads that can process text, images, and videos to generate text. Customers can adjust how much step-by-step “thinking” the model performs before responding, balancing intelligence depth with speed and cost—ideal for customer service chatbots, document processing, and business automation. Nova 2 Lite delivers industry-leading price performance in its intelligence class. It is equal or better on 13 out of 15 benchmarks compared to Claude Haiku 4.5, equal or better on 11 out of 17 benchmarks compared to GPT-5 Mini, and equal or better on 14 out of 18 benchmarks compared to Gemini Flash 2.5. Nova 2 Lite demonstrates standout capabilities in processing documents, extracting information from videos, generating code, providing accurate grounded answers, and automating multi-step agentic workflows.

Nova 2 Pro is Amazon’s most intelligent reasoning model that can process text, images, video, and speech to generate text. It’s ideal for highly complex tasks like agentic coding, long-range planning, and sophisticated problem-solving—where the highest accuracy is essential. The model can also serve as a “teacher” for knowledge distillation—transferring its capabilities into smaller, more efficient “student” models for specific domains and use cases. Nova 2 Pro is equal or better on 10 out of 16 benchmarks compared to Claude Sonnet 4.5, equal or better on 8 out of 16 benchmarks compared to GPT-5.1, equal or better on 15 out of 19 benchmarks compared to Gemini 2.5 Pro, and equal or better on 8 out of 18 benchmarks compared to Gemini 3 Pro Preview. Nova 2 Pro demonstrates strengths in multi-document analysis, video reasoning, following complex instructions, solving advanced math, and executing agentic and software engineering tasks.

Both Nova 2 Lite and Nova 2 Pro have built-in web grounding and code execution capabilities, meaning they can search the web for current information and run code directly—ensuring responses stay anchored in up-to-date facts rather than relying solely on training data.

Nova 2 Sonic is Amazon’s speech-to-speech model that unifies text and speech understanding and generation for real-time, human-like conversational AI. It features expanded multilingual support with expressive voices, higher accuracy, and a one-million token context window for sustained interactions, while enabling seamless switching between voice and text. The model handles tasks asynchronously, letting users continue natural conversations—even switching topics—while actions like booking flights complete in the background. It also seamlessly integrates with Amazon Connect, telephony providers (Vonage, Twilio, AudioCodes), and conversational AI frameworks (LiveKit, Pipecat), making it ideal for customer service applications, AI assistants, and interactive voice experiences. Nova 2 Sonic offers industry-leading price performance and quality compared to OpenAI’s gpt-realtime and Gemini 2.5 Flash models available through their realtime APIs.

Nova 2 Omni is a unified multimodal reasoning and generation model that can process text, images, video, and speech inputs while generating both text and images—an industry first. It handles up to 750,000 words, hours of audio, long videos, and hundred-page documents, simultaneously analyzing entire product catalogs, testimonials, brand guidelines, and video libraries at once. This eliminates the cost and complexity of connecting multiple specialized models. For example, marketing teams can analyze product details across all formats to instantly generate complete campaigns including headlines, copy, social posts, and visuals in one workflow. While there are no comparable models in the industry to Nova 2 Omni, it demonstrates strengths in public benchmarks of multimodal reasoning on documents, images, videos, and audio, and can generate high-quality images similar to other leading image-generation models.

Organizations like Cisco, Siemens, Sumo Logic, and Trellix are using Nova 2 models for applications ranging from agentic threat detection to video understanding and voice AI assistants.

Organizations embedding proprietary knowledge into AI applications currently face three compromises: customizing proprietary models in ways that only scratch the surface for integrating an organization’s expertise, continuing to train open-weights models without access to the original training data, which risks the model regressing on foundational capabilities like instruction following, or building from scratch at enormous expense. What organizations need is access to both frontier model capabilities and the ability to deeply integrate their expertise.

Nova Forge empowers organizations to build their own optimized variants of Nova—we call them “Novellas”—by blending their proprietary data with Nova’s frontier capabilities. The service pioneers “open training”—giving exclusive access to pre-trained, mid-trained, and post-trained Nova model checkpoints so customers can mix their proprietary data with Amazon Nova-curated datasets at every stage of model training. The result is a customized model that combines Nova’s full knowledge and reasoning power with deep understanding of each organization’s specific business. Customers can start building their own Novellas with Nova 2 Lite today. In addition, Nova Forge customers get early access to Nova 2 Pro and Nova 2 Omni, which gives them a head start in building applications and their Novellas with even more capable Nova models.

Beyond model checkpoints and data-mixing capabilities, Nova Forge offers three additional powerful capabilities: First, the ability to train AI using your own environments, which are referred to as reinforcement learning “gyms.” These gyms are synthetic environments where the models learn from simulated scenarios that reflect their real-world use cases. Second, option to create smaller, faster models that maintain their intelligence at a lower cost—trained on AI-generated examples from larger models through a process called synthetic data-based distillation. Third, access to a responsible AI toolkit that allows them to implement safety controls.

Organizations like Booking.com, Cosine AI, Nimbus Therapeutics, Nomura Research Institute, OpenBabylon, Reddit, and Sony are building their own models with Nova Forge to better serve their unique requirements.

“Working with Nova Forge is allowing us to improve content moderation on Reddit with a more unified system that’s already delivering impressive results,” said Chris Slowe, CTO, Reddit. “We’re replacing a number of different models with a single, more accurate solution that makes moderation more efficient. The ability to replace multiple specialized 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.”

Once customers create their own frontier model with Nova Forge, they can deploy it on Amazon Bedrock with the same enterprise-grade security, scalability, and data privacy as all other Bedrock models. This complete solution—from building their own frontier model to production deployment—ensures organizations achieve optimal AI performance tailored to their specific business needs, with exclusive use of their model securely hosted on AWS.

Nova Act is now available as a service on AWS for building and deploying highly reliable AI agents that can take actions in web browsers. Powered by a custom Nova 2 Lite model, Nova Act provides the fastest and easiest path to build and manage fleets of agents that automate browser-based tasks. Nova Act delivers 90% reliability on early customer workflows and outperforms competing models on relevant benchmarks.

Nova Act achieves breakthrough reliability by training a custom Nova 2 Lite model through reinforcement learning, running thousands of tasks on hundreds of simulated web environments. This style of training allows Nova Act to excel on UI-based workflows like updating data in a customer relationship management (CRM) system, testing website functionality, or submitting health insurance claims.

With Nova Act, developers can start prototyping an agent in minutes with a no-code playground using natural language prompts, refine that Nova Act agent in familiar IDEs like VS Code, and then deploy to AWS. What customers build and test locally scales in production, with comprehensive management tools and monitoring through the Nova Act AWS console.

Organizations across sectors are already seeing results with Nova Act:

To learn more about Amazon Nova, visit aws.amazon.com/bedrock/nova. Start building for free at nova.amazon.com/dev.

Get the latest news from AWS re:Invent, including all things agentic and generative AI, product and service announcements, and more.