Introducing AWS Nova 2: Advancing Custom AI Models for Enterprise Innovation

Amazon Web Services (AWS) is intensifying its focus on tailored artificial intelligence solutions with the launch of a new suite of models and a groundbreaking custom model service. These innovations empower businesses to infuse their proprietary data and domain expertise into AI systems, enhancing personalization and operational intelligence.

Expanding the Nova Family: Enhanced AI Models for Diverse Applications

Building on the success of its initial Nova models, AWS has unveiled the Nova 2 series, which includes Nova 2 Lite, Nova 2 Pro, Nova 2 Sonic, and Nova 2 Omni. Each model is designed to address specific enterprise needs, ranging from everyday reasoning tasks to complex problem-solving and multimodal data processing.

• Nova 2 Lite: A cost-efficient, high-speed reasoning model optimized for routine tasks. It processes text, images, and videos to generate coherent textual outputs, making it ideal for scalable applications.
• Nova 2 Pro: AWS’s most advanced reasoning model, capable of handling intricate challenges such as coding automation, strategic planning, and sophisticated problem-solving. It also serves as a “teacher” model for knowledge distillation projects.
• Nova 2 Sonic: A multilingual speech-to-speech model that delivers natural, fluid conversations with enhanced voice expressiveness and a 1-million-token context window, enabling seamless topic transitions.
• Nova 2 Omni: A powerful multimodal model that simultaneously analyzes extensive text, audio, video, and image data-processing up to 750,000 words and hours of multimedia content-to generate high-quality images and insights.

Nova Forge: Revolutionizing Custom AI Model Development

At the heart of AWS’s latest AI offerings is Nova Forge, a pioneering service that allows enterprises to customize pre-trained Nova models by integrating their unique datasets without compromising the foundational knowledge of the model. This approach addresses a critical challenge in AI development: maintaining core capabilities while adapting to specialized domains.

Nova Forge utilizes an “open training” methodology, blending proprietary data with Amazon-curated datasets throughout the training process. This ensures that the resulting custom models-referred to as “Novellas”-retain robust reasoning and instruction-following abilities while being finely tuned to specific business contexts.

Currently, customization is available for Nova 2 Lite, with plans to extend support to other Nova 2 variants. Additionally, Nova Forge offers “reinforcement learning gyms,” enabling organizations to simulate real-world environments for training AI agents, fostering faster, more efficient models equipped with responsible AI toolkits.

Real-World Application: Reddit’s Moderation AI

One notable example of Nova Forge’s impact is Reddit’s use of the platform to embed community-specific knowledge into a moderation AI system. This tailored model enhances content oversight by understanding the nuances of Reddit’s diverse user base, demonstrating the practical benefits of customized AI solutions.

Performance and Industry Positioning of Nova 2 Models

With tens of thousands of organizations already leveraging the original Nova models, AWS anticipates similar widespread adoption for the Nova 2 series. According to AWS CEO Matt Garman, Nova 2 Lite is expected to become the “workhorse” for many enterprises due to its balance of performance and cost-efficiency, while Nova 2 Pro is suited for more demanding, complex tasks.

Independent benchmark tests reveal that Nova 2 Lite matches or surpasses competitors such as Claude Haiku 4.5, GPT-5 Mini, and Gemini Flash 2.5 across a majority of evaluated metrics. Nova 2 Pro similarly excels in multi-document analysis, video reasoning, advanced mathematics, and agentic engineering, outperforming models like Claude Sonnet 4.5, GPT-5.1, and Gemini 2.5 Pro.

Both Nova 2 Lite and Pro incorporate advanced grounding and code execution features, enhancing their utility in real-world applications.

Multimodal and Speech Capabilities: Nova 2 Sonic and Omni

Nova 2 Sonic advances speech-to-speech AI with support for multiple languages, improved voice naturalness, and the ability to maintain coherent conversations even when topics shift abruptly. This makes it highly suitable for customer service, virtual assistants, and interactive voice applications.

Meanwhile, Nova 2 Omni stands out as a unique multimodal powerhouse, capable of processing vast amounts of text, audio, video, and images simultaneously. It excels in analyzing complex datasets such as product catalogs, customer testimonials, brand assets, and extensive video libraries, generating high-quality images and insights that rival leading image-generation models.

Future Outlook and Availability

Currently, Nova Forge supports only AWS’s proprietary Nova models, with no immediate plans to integrate third-party open-source models hosted on Amazon Bedrock. However, the platform’s modular design and robust capabilities position it as a key enabler for enterprises seeking to harness AI tailored to their unique operational landscapes.