Amazon Has New Frontier AI Models—and a Way for Customers to Build Their Own

Summary

Amazon unveiled the second generation of its Nova family at re:Invent: Nova 2 Lite and Nova 2 Pro (LLMs), Nova Sonic (real-time voice), and Nova Omni (fully multimodal simulated reasoning). The headline feature is Nova Forge — a tool that lets AWS customers add their own data during multiple stages of model training, including custom pretraining, to produce specialised ‘frontier’ models. Early testers such as Reddit, Booking.com, Sony and Nimbus Therapeutics have used Nova Forge to create domain-expert models (for example, a Reddit moderation specialist). Amazon positions Nova 2 Pro as competitive with the latest models from OpenAI, Google and Anthropic, and emphasises that Nova Forge reduces the cost and barriers to building tailored frontier models — albeit locked to Amazon’s cloud.

Key Points

Amazon announced Nova 2 Lite, Nova 2 Pro, Nova Sonic (voice) and Nova Omni (multimodal) at re:Invent.
Nova Forge allows customers to add training data at several training stages, including custom pretraining normally reserved for labs.
Reddit used Nova Forge plus fine-tuning to build a Reddit-specific moderation model that can handle content other models refuse to process.
Amazon claims Nova 2 Pro matches or exceeds benchmarks for GPT-5/GPT-5.1, Gemini Pro 2.5/3.0 Pro and Anthropic’s Sonnet 4.5 on a range of tasks.
Nova Omni is presented as a fully multimodal simulated-reasoning model accepting text, images, audio and video as inputs.
Nova Forge is cloud-locked to AWS — it lowers cost vs building from scratch but ties customers into Amazon’s infrastructure.
Early enterprise testers include Booking.com, Sony and a biotech (Nimbus Therapeutics), indicating cross-industry interest.

Content Summary

Amazon’s Nova 2 lineup expands its frontier-model offerings with both large language and multimodal systems. The major technical and commercial differentiator is Nova Forge: a service that hands customers unfinished base models that they can further train with their own data at different stages, including the base-pretraining phase. This is more flexible than conventional fine-tuning and aims to produce domain-expert models more efficiently and cheaply than building a model from scratch. Amazon says Nova 2 Pro performs on par with top-tier competitors and touts agentic capabilities, while Nova Omni demonstrates advanced multimodal reasoning.

The approach has practical appeal: Reddit says it built a specialist model that understands the nuance of its platform better than off-the-shelf LLMs. But Nova Forge’s benefits come with caveats — customers must run on AWS, and Amazon has not disclosed exact pricing or cost savings. The move fits Amazon’s broader strategy to capture cloud AI workloads and to offer enterprises tighter integration between their data, specialised models and cloud infrastructure.

Context and Relevance

This matters because companies increasingly want models that behave like domain experts rather than generic assistants. Many organisations struggle with adapting general-purpose models to specialised tasks; Nova Forge promises a middle ground by enabling deeper customisation without the prohibitive cost of building a frontier model from scratch. For AWS customers, this could speed production deployments of tailored AI in areas such as content moderation, search, customer support, and industry‑specific workflows.

On the industry level, Amazon’s announcement intensifies the competition among cloud providers and AI labs (Google, Microsoft, OpenAI, Anthropic, Nvidia). It also highlights two tensions: (1) the commercial trade-off between convenience and vendor lock-in, and (2) the rising importance of multimodal, agentic capabilities in practical applications.

Why should I read this?

Short version: if your organisation uses or plans to use cloud AI, skim this — Nova Forge could save you time and money when you want a model that actually understands your data. It’s not just another LLM release; it’s a different way to make models behave like real experts for specific business tasks. If you’re non-technical, it still matters because this is how companies will start shipping smarter, more useful AI in production.

Author Style

Punchy — this is presented as a meaningful step for enterprise AI: the article spells out technical novelty, business use cases and who’s already testing it, so readers can judge impact quickly.