AWS Nova 2 & Nova Forge: Custom AI for enterprises

Amazon Web Services has taken a bold leap in the generative AI race with the launch of its Nova 2 series at re:Invent 2025. The new lineup, designed for enterprise-scale tasks, introduces four advanced models, Nova Lite, Nova Pro, Nova Sonic, and Nova Omni, alongside a groundbreaking service called Nova Forge, empowering organizations to build custom AI models from the ground up.

The Nova 2 model family: Purpose-built for enterprise

The Nova 2 series marks a significant evolution in AWS’s AI strategy, offering a suite of models tailored for diverse business needs:

• Nova Lite: A lightweight, multimodal model optimized for speed and cost, ideal for text generation and basic reasoning tasks.
• Nova Pro: Engineered for complex, multi-step reasoning, long-range planning, and advanced agentic workflows. It excels in scenarios like multi-document analysis and software migrations.
• Nova Sonic: Specialized for speech dialogue, enabling natural and responsive voice interactions.
• Nova Omni: Boasts a massive context length of up to 750,000 words, making it suitable for processing lengthy documents and large-scale data.

These models are designed to address common enterprise challenges such as cost, latency, and integration complexity, offering industry-leading price-performance ratios. For example, Nova Lite is priced at $0.045 per million tokens for input and $0.09 for output, about half the cost of comparable models.

Nova Forge: Custom AI model building at scale

One of the most anticipated announcements was Nova Forge, a service that allows customers to participate in the pre-training, mid-training, or post-training stages of model development. For an annual fee of $100,000, organizations can inject private data to build custom models from scratch, ensuring unique business requirements are met without compromising on reasoning ability.

Early adopters like Reddit have already leveraged Nova Forge to train content moderation models, reportedly outperforming commercially available large models in internal evaluations. AWS also offers additional consulting support for enterprises needing hands-on assistance from Amazon engineers.

Enterprise applications and industry adoption

The Nova 2 series is already making waves in real-world applications. Internal teams at Amazon, including those behind e-commerce and Alexa, have adopted Nova Forge for business-scale model development. External customers are using the models for tasks ranging from code modernization to infrastructure transformation.

For instance, AWS Transform, a multi-agent, multi-user system, uses Nova 2 Lite to break down complex user requests into agent and tool calls, improving tool-calling efficiency for code modernization by up to 60%.

Future developments and roader capabilities

Amazon is not stopping here. The company plans to expand the Nova family with additional models, including a speech-to-speech model in Q1 2025 and an “any-to-any” multimodal model by mid-2025. The Nova models also support over 200 languages and offer extensive context length capabilities, with future releases aiming for context lengths exceeding 2 million input tokens.

Integration with other cloud platforms is also on the horizon, as AWS announced an interconnection agreement with Google Cloud, enabling easier access to competing models like Gemini across platforms.

Conclusion

The Nova 2 series and Nova Forge represent a major step forward in AWS’s commitment to delivering enterprise-grade AI solutions. By focusing on cost-effectiveness, customization, and scalability, Amazon is empowering organizations to harness the full potential of generative AI for their unique business challenges. As the AI landscape continues to evolve, AWS’s latest innovations position it as a leader in the race for intelligent, adaptable, and affordable enterprise AI.