AWS Nova Forge: Open-Training Platform Enables Deep Customization of Nova 2 Models

AWS Nova Forge is presented as an open-training platform that enables deep customization of Nova 2 models, expanding AWS’s Nova family with Nova 2 Lite, Nova 2 Pro, Nova 2 Sonic, Nova 2 Omni and Nova Act, as well as the Nova Forge framework for bespoke knowledge models. In an interview during re:Invent 2025, according to Eshan Bhatnagar, Amazon’s Director of AGI Products, AWS is emphasising how Nova Forge helps enterprises reflect their operational realities through more flexible model customization.

The platform supports fine-tuning at pre-training, mid-training or post-training stages, allowing domain-specific expertise to be woven directly into the model rather than treated as an external attachment, which can previously lead to inconsistent results. AWS also continues to enhance its guardrails framework so customers can define enforceable behavioural boundaries.

Regarding model selection, AWS positions itself as a provider of choice rather than prescription, with customers free to adopt the latest models or rely on the evolving Nova family. Bhatnagar also indicated that Nova Forge could be applied to Alexa services, enabling targeted tuning without changing the underlying Nova model, though there are no plans to introduce full Chinese-language support to Alexa+.