Amazon Nova Act Review: Can AWS Finally Compete with OpenAI and Google?

Amazon Nova Act Review: Can AWS Finally Compete with OpenAI and Google? On December 1, 2025, during the AWS re:Invent conference, Amazon unveiled Nova 2. While the industry was focused on the battle between OpenAI and Google, Amazon prepared a massive response: not just a single model, but a complete family designed for the enterprise. The main selling point of Nova Act: Amazon promises performance matching the competition at a cost 75% lower. Let’s look exactly at what Nova offers and if these promises hold up against the reality of the market. ➥ What is Amazon Nova Act? Amazon Nova is AWS’s direct answer to GPT‑5, Claude, and Gemini. Amazon’s approach differs slightly by offering a range of models integrated into Amazon Bedrock . The idea is to offer a solution for every budget and task type, all secured by the AWS infrastructure that major enterprises already use. ➥ The Nova 2 Family in Detail Nova 2 Lite: Speed and Economy This model is designed to process high volumes of requests at a lower cost. Multimodal, it handles text, images, and video. It supports over 200 languages and offers a 300,000-token context window. It is ideal for chatbots, content moderation, or real-time translation. Nova 2 Pro: The Versatile Standard This is the balanced model between intelligence and price. It processes text, image, video, and audio. Its strengths include video summarization, mathematical reasoning, and software development. An interesting feature is “adjustable reasoning depth,” which allows you to define how long the model should think before responding, influencing both cost and speed. It also includes web access and code execution. Nova 2 Sonic: The Real-Time Voice Assistant Nova 2 Sonic aims to compete with advanced voice models. It offers state-of-the-art speech understanding and generation with a bidirectional streaming API for fluid conversations. It features a one-million token context window and allows for instant switching between voice and text modes. Nova 2 Omni: The Experimental Model Still in the experimental phase, this model offers unified multimodal generation. It is designed for creative tasks requiring total fluidity between image, audio, video, and text. ➥ The Real Innovations: Forge and Act Beyond standard models, Amazon presented two tools targeting specific enterprise needs. Nova Forge: Custom Training Nova Forge is an open training service allowing companies to create their own versions of Nova models using their proprietary data. Instead of using a generic model, you can intervene in the middle of the training process. The starting investment is significant (starting at $100,000 per year), making it suitable for enterprises with highly specific needs, such as medical diagnostics or detailed financial analysis. Amazon Nova Act: Reliable Agents? Nova Act tackles the major problem with autonomous AI agents: their failure rate. With Nova Act, Amazon announces 90% reliability on browser-based automation workflows. This allows for automating complex tasks like cross-system data entry or navigating multiple client portals, transforming experimental technology into a deployable solution. ➥ Performance and Pricing According to the benchmarks presented, Nova 2 Pro matches or exceeds GPT‑5 and Gemini on most reasoning tasks. Nova Premier (the most advanced model) excels in information retrieval and visual reasoning. However, the real asset remains the pricing. Amazon announces costs 75% lower than the competition. For example, Nova Premier is listed at $2.50/million tokens input and $12.50 output. For a company spending millions on AI infrastructure, this price difference is a major factor. ➥ Pros and Cons Cost: The 75% reduction compared to competitors makes AI more accessible. AWS Ecosystem: Security and compliance inherited from Bedrock provide reassurance for CIOs. Agent Reliability: Nova Act’s 90% success rate is promising. Customization: Nova Forge allows for deep adaptation of models. Context: From 300,000 to 1 million tokens to manage heavy workloads. Vendor Lock-in: Nova only runs on Amazon Bedrock. Using Nova means committing further to AWS. Product Maturity: Unlike GPT‑5, Nova has not yet been proven over time by the general public. Complexity: Implementation requires knowledge of the AWS environment. ➥ Verdict: Is It Worth It? If your company already uses AWS, the question barely needs asking. Native integration, combined with drastic cost reductions and features like Nova Act, makes Nova a compelling option. For others, the choice is more complex. The “lock-in” with AWS is a real barrier. However, if you need highly customized models via Forge or reliable autonomous agents, Amazon is offering tools here that the competition still struggles to match operationally. Amazon Nova Act might not be an absolute technological revolution, but it is a pragmatic and economic response to enterprise needs. Visit Amazon Nova website