Hold onto your hats, folks! Amazon is shaking up the AI world with its new frontier models, and they're giving you the keys to build your own. This is a game-changer, and here's why…
At the re:Invent conference in Las Vegas, the e-commerce giant unveiled the second generation of its Nova AI models. While they might not be household names like OpenAI or Google's offerings just yet, Amazon's strategy of making them highly customizable could quickly change that, especially among its cloud users.
Let's break down what's new: Amazon is rolling out two improved large language models, Nova Lite and Nova Pro. They're also introducing a real-time voice model called Nova Sonic, and an experimental model, Nova Omni, which is designed to perform reasoning using a combination of images, audio, video, and text. These models are available now, though access is currently limited.
But the real innovation? Amazon's new tool, Nova Forge. This allows customers to create specialized frontier models by adding their own training data to unfinished versions of the Nova 2 Lite and Pro models.
You're probably thinking, "Wait, isn't fine-tuning AI models already a thing?" Yes, you can fine-tune existing models like Google's Gemini or OpenAI's GPT. But here's where it gets controversial... Amazon's approach lets you add data during various stages of model training, including the initial base model creation – a process usually reserved for the big AI labs.
"Everyone is looking for a frontier model that's an expert in their domain," says Rohit Prasad, who leads Amazon’s AI efforts. He explains that Nova Forge was developed to empower internal teams, including those behind Alexa, to build custom models. He calls it a "new open training paradigm."
One of the first to test this approach is Reddit. They used Nova Forge to build a custom model designed to identify content that violates the platform's rules.
According to Reddit's CTO, Chris Slowe, fine-tuning a conventional model wouldn't cut it. Most models are designed to avoid offensive content entirely, which means they would struggle to analyze certain materials. Custom pre-training, combined with fine-tuning, resulted in a model that truly understands Reddit.
"Other LLMs understand Reddit as a concept, and how Reddit works, but they're not down in the weeds," Slowe explains. "We really built a Reddit expert model." He adds that this customized model has many potential uses, starting with automating content moderation.
Other companies, like Booking.com, Sony, and Nimbus Therapeutics, are also testing Nova Forge.
And this is the part most people miss... This move could be brilliant, as companies seek AI tools that go beyond the capabilities of current general-purpose models. A recent survey revealed that about three-quarters of U.S. companies consider AI a high priority. However, they also report challenges, including a lack of expertise and resources to build custom models.
Today, most AI models are either closed (accessed via API or app) or open (downloadable and run on your hardware). Many companies opt for open models, especially those from Chinese companies like Alibaba and DeepSeek, because they're cheaper to experiment with and easier to modify. However, the data used to train these open models isn't usually released, which complicates the tuning process.
Nova Forge offers a new approach, although it's tied to Amazon's cloud. Building a large language model from scratch can cost tens or even hundreds of millions of dollars. Prasad suggests that a frontier model built using Nova Forge should be significantly cheaper, though he didn't provide specific figures.
Amazon is still something of a dark horse in the AI race, having been a latecomer to developing truly cutting-edge AI language models. However, the company is quietly building a portfolio of advanced AI capabilities. For instance, they've integrated generative AI into their shopping platform with a chatbot helper called Rufus.
Like other tech giants, Amazon is investing billions in AI infrastructure. This is a massive bet that the demand for AI will continue to grow rapidly.
Amazon is competing with Google and Microsoft for cloud customers. OpenAI is also rapidly building its own infrastructure and could become a cloud player itself. Amazon is also looking to challenge Nvidia’s hardware dominance.
Amazon claims Nova 2 Pro matches or exceeds OpenAI’s GPT-5 and GPT-5.1, Google’s Gemini Pro 2.5 and Gemini 3.0 Pro, and Sonnet 4.5 from Anthropic across a range of benchmarks. Prasad notes that the model is especially good at agentic tasks like following complex instructions and using tools on a computer. The company says that its smaller model, Nova 2 Lite, is similar to Claude 4.5 Haiku, GPT-5 Mini, and Gemini Flash 2.5 on various benchmarks.
Nova 2 Omni shows that Amazon is no slouch in AI research these days. A fully multimodal reasoning model, it can take images, audio, and video as well as text as input and can perform simulated reasoning to generate output. Prasad says to his knowledge no other AI company has released a fully mutli-model of this kind.
Reddit’s Slowe says the customizable nature of Nova is probably its most important quality. “I do believe it has a lot of potential,” he says. “For a large set of situations, it will be substantially better than what we get off the shelf.”
So, what do you think? Is Amazon onto something big with Nova Forge? Will this level the playing field for smaller companies wanting to build custom AI? Share your thoughts in the comments below! Do you agree with the potential of Amazon's new approach, or do you have reservations?
