Rohit Prasad

As I close an incredible chapter at Amazon, I'm deeply grateful for the journey and people who made it so meaningful. I joined Amazon 12 years ago to work on Alexa because it felt like one of those ideas that bordered on impossible – a "Star Trek" dream that was absolutely worth building for customers. What followed exceeded anything I could have imagined. Together, we built conversational AI that touches the daily lives of hundreds of millions of households around the world. In more recent years, building Amazon Nova’s frontier models and services has been just as exhilarating. Beyond the technology itself, I’m most proud of the foundations we’ve laid together. Seeing the energy around Nova 2, Nova Forge, and Nova Act at re:Invent earlier this month was a highlight I'll always treasure – and in many ways, it felt like the perfect capstone. Amazon has taught me a lot, and I’ve grown here in ways that would have been impossible elsewhere. The culture – brilliant people at every level, deeply obsessed with customers and willing to take long-term bets – is truly unmatched. But above all, I’m grateful for the friendships and partnerships with teammates who made the "impossible" worth building…and so much fun :-) With these foundations now in place, I felt it was the right moment for me to begin a new adventure. The team is in great hands with Peter DeSantis, and I have no doubt that bringing foundational AI, silicon, and quantum together will be a force multiplier. To my teammates, partners, and customers: thank you for sharing this journey with me. I'm excited for what comes next and I can’t wait to see the extraordinary things you build!

Since launching Amazon Nova Forge, we’ve had great conversations with the community about what it can do and why we built it. I explore the technical challenges we navigated to make "open training" a reality here: amazon.science/blog/amazon-no… We built Forge initially because our own internal teams needed it. Across Amazon's diverse businesses, teams needed models with deep expertise in their specific domains – and our external customers wanted the same. There were two key challenges to overcome. First, how can massive amounts of domain-specific data be infused into the model without catastrophic forgetting of its foundational capabilities? For example, if you have specialized data for drug discovery, can you incorporate it during early training stages like pre-training and mid-training? Second, how can we make the toolchain and recipes self-serve for customers? This required solving a collection of science and engineering problems – from data engineering (what data, in what proportions, at what stages), learning dynamics (how to keep training predictable as the data distribution shifts), and reliable scaling on Amazon SageMaker AI. We spent months refining these recipes with internal teams, then validated them with early customers to make each training stage accessible and predictable. Each time a customer used Forge recipes, their model significantly outperformed other leading alternatives. This is just the beginning of what's possible when we lower the barriers to frontier model development. Thank you to the partners who worked with us from design to general availability. Excited to see what organizations build through this "open training" paradigm.

🚀 The Amazon Nova 2 family is here — our next generation of foundation models that deliver industry-leading price-performance across reasoning, multimodal processing, conversational AI, code generation, and agentic tasks. We're launching four new models today: Nova 2 Lite - Our fast, cost-effective reasoning for everyday workloads Nova 2 Pro - Our most intelligent reasoning model for complex tasks Nova 2 Omni - The industry's first reasoning model that processes text, images, video, and speech inputs while natively generating both text and images Nova 2 Sonic - Industry-leading speech-to-speech model for real-time conversational AI What makes Nova 2 special? 🤔 ⚡ Adjustable Thinking Budget: Toggle reasoning on/off, choose low/medium/high thinking levels based on your task complexity—from fast chatbot responses to sophisticated multi-step agentic workflows 🎯 Production-Ready Performance: Nova 2 Lite delivers 7x lower cost and 5x faster responses than Nova Premier (our previous flagship), while outperforming it on multi-step problem solving and agentic tasks 🌍 True Multimodal with Long Context: Process text, images, video, and speech with 1M token context—handle 400+ page documents, 90-minute videos, or massive codebases in a single prompt 🎨 Unified Generation: Nova 2 Omni is our first reasoning model that understands multimodal inputs AND natively generates text and images—enabling use cases like marketing content creation, customer support call transcription, video analysis, document understanding, and natural language image editing 🗣️ Real-Time Voice: Nova 2 Sonic delivers natural, expressive conversations with asynchronous tool calling and seamless voice-text switching 🤖 Built for Agents: Native support for tools, APIs, and MCP servers. You can see the demo below where we used Nova 2 with AgentCore and Strands to automatically update the LangChain repository for this launch—it analyzed the code, created a branch, implemented changes, and submitted a PR ready to merge 🛡️ Responsible AI: Built-in safety controls and content moderation with advanced robustness to adversarial attacks Customers are already having great results and we are excited to see what developers build with Nova 2! Nova 2 Lite and Nova 2 Sonic are available today. Nova 2 Pro and Nova 2 Omni are in preview for Nova Forge customers. Start building: lnkd.in/e8EErNxB #AmazonNova #Nova2 #ReasoningModels #GenerativeAI #AIAgents #AWS #AmazonBedrock #MachineLearning

Amazon is back with Nova 2.0, a substantial upgrade over prior Amazon Nova models and demonstrating particular strength in agentic capabilities Amazon has released Nova 2.0 Pro (Preview), its new flagship model; Nova 2.0 Lite, focused on speed and lower cost; and Nova 2.0 Omni, a multimodal model handling text, image, video and speech inputs with text and image outputs. Key benchmarking takeaways: Amazon back amongst top AI players: This is Amazon’s latest release since Nova Premier and Amazon’s first release of reasoning models. Nova 2.0 Pro jumps 30 points in the Artificial Analysis Intelligence Index over Premier and Lite 38 points. This represents a huge increase in capabilities and Amazon’s return to being amongst the top AI players. Strengths in agentic capabilities: Agentic capabilities including tool calling is a strength of the models, Nova 2.0 Pro scores 93% on τ²-Bench Telecom and 80% on IFBench on medium and high reasoning budgets respectively (complete benchmarks for high reasoning coming soon). This places Nova 2.0 Pro Preview amongst the leading models in these benchmarks. Multimodal: Nova 2.0 Omni is one of few models, alongside most notably the Gemini model series, that can natively handle text, image, video and speech inputs. This is a new differentiator for Amazon’s Nova model series. Competitive pricing: Amazon has priced Nova 2.0 Pro at $1.25/$10 per million input/output tokens, and considering token usage the model took $662 to run our Artificial Analysis Intelligence Index. This is substantially less than other frontier models like Claude 4.5 Sonnet ($817) and Gemini 3 Pro ($1201), but remains above others including Kimi K2 Thinking ($380). Nova 2.0 Lite and Omni are both priced at $0.3/$2.5 per million input/output tokens. See below for further analysis

We’re introducing a big upgrade to Amazon Nova Act. This is our first model with large-scale RL for web tasks. We achieve a frontier-class browser agent that's also the most cost-effective. Plus, seeing your work at a keynote is an insane feeling :) More on the RL behind it🧵

Excited to share what our lab has been baking: Amazon Nova Act! Trained with large scale RL on diverse web gyms, Nova Act achieves SOTA on multiple public web agent benchmarks. Check it out!🚀 labs.amazon.science/blog/amazon-no…

Following the success of the Amazon Nova Act SDK, I’m thrilled to share that Nova Act is now available as a service on AWS for building and managing highly reliable AI agents at scale! Our team has made giant strides on automating complex, business-critical workflows. Early enterprise customers are seeing 90% reliability on UI-based tasks – the key threshold where organizations can actually trust technology to handle mission-critical work. We achieved this by training Nova Act as a unified, end-to-end system through reinforcement learning, instead of bolting together separate components. Crucially, we also instilled a sense of humility. Nova Act knows when it's stuck and proactively asks for human help. Results from customers like Hertz, Sola, and Amazon Leo validate this approach. I’ve always believed AI should handle the tedious tasks we don't enjoy. Nova Act is reliable (and humble) enough to do that at scale. If you're ready to put agents to work, check it out: nova.amazon.com/act

Really enjoyed Matt’s keynote at #AWSreInvent today. So much innovation happening in @awscloud, and you could see it with the array of launches he unveiled. So many parts of the keynote worth watching, but will point to a few: 1/ Excited about the availability of Trainium3. Trainium2 has substantial traction, is a multi-billion-dollar revenue run rate business, has 1M+ chips in production, and 100K+ companies using it as the majority of Bedrock usage today. Trainium2 has price-performance advantages over other GPU options that are compelling, and Trainium3 will deliver at least 4.4x more compute performance, 4x greater energy efficiency, and almost 4x more memory bandwidth than Trainium2: youtu.be/q3Sb9PemsSo?si… 2/ Worth double clicking on AgentCore, which has changed the security and scalability of deploying agents into production. AgentCore is a set of flexible building blocks that can be used in any combination developers want, and AWS added two more in Policy and Evaluations. AgentCore has a lot of momentum: youtu.be/q3Sb9PemsSo?si… 3/ Nova Forge is a game-changer for companies wanting to customize a frontier model with their own proprietary data. Like equipping a young person with a better knowledge foundation to keep learning, LLMs are better able to solve problems and improve if they’re trained early on with companies’ differentiated data. To do so, companies need earlier versions of the frontier model and ability to mix their own data with the model’s data. This is what Forge provides and this “open training” allows companies to develop Novellas that are their own, optimized versions of Nova they can use for their AI apps and agents. Customers have been itching for this sort of capability, and Forge is a uniquely compelling approach: youtu.be/q3Sb9PemsSo?si… 4/ Agents will become the primary way companies get value from AI. We have built some compelling agents for our customers in Kiro (for coding), Quick (for knowledge workers to leverage their own data, analytics, and routines), Transform (to migrate from one software source to another), and Connect (call center agents). But there are tasks customers want agents to solve more autonomously and over longer durations, and new AWS frontier agents—Kiro autonomous agent, AWS DevOps Agent, and AWS Security Agent—are exciting: youtu.be/q3Sb9PemsSo?si… 5/ Finally, I enjoyed Matt’s ending 25 launches in 10 mins, both because it was action-packed and represented so much useful delivery. The reason so many people cheered these launches is because even though they’re less sexy, they’re the meat and potatoes core infrastructure needs customers have—and with so much of the total cloud infrastructure running on top of AWS, these launches will make a lot of people’s lives easier and better every day: youtu.be/q3Sb9PemsSo?si… Enjoy (and just day 1 of announcements today for AWS :-)!

Early results from our internal teams and customers like Reddit, Sony, Nimbus Therapeutics, and Nomura Research have been remarkable. Nova Forge is a major step forward in democratizing AI for every organization. Huge congrats to the team for making this real. Come build your own frontier model with us! aboutamazon.com/news/aws/aws-a…

Nova Forge introduces a new paradigm of "open training" for building your own expert models. You get access to checkpoints at every stage of model development (pre-trained, mid-trained, post-trained) and can blend your proprietary data with Amazon Nova-curated data. When you blend your data into early training stages, the model learns to reason about your domain as an expert. The result? Your own frontier model – your own "Novella" as we like to call it. (You can name it whatever you want.) 😉

We've been hard at work solving a persistent industry problem: frontier models launch with impressive benchmarks, organizations test them, then they don't work for actual needs. To help bridge this gap, we're excited to announce Amazon Nova Forge – a new way to build frontier AI models that are experts in your domain.

If you're attending NeurIPS, stop by our booth (#1523) this week. We'll be hosting demos, info sessions, and meet and greets with our team. 📆 Full schedule: amzn.to/48rCLaj #NeurIPS2025

Applications are open for the #AmazonNovaAIChallenge. University teams are invited to advance trusted, useful agentic AI for software engineering: amazon.science/nova-ai-challe…

Security researchers are critical partners in building safer AI, pushing our systems in new ways to uncover vulnerabilities and edge cases that make our models stronger. That's why we're launching a private bug bounty program for Amazon Nova models – working with researchers and universities to find vulnerabilities like prompt injection and misuse scenarios, with rewards up to $25K for validated findings. amazon.science/news/amazon-la…

Amazon Nova Web Grounding is now generally available – a turnkey RAG solution that helps developers build more accurate AI applications with cited sources. No complex pipelines to manage, just reliable, current information when you need it. More here: aws.amazon.com/blogs/aws/buil…

Year 2 of the Amazon Nova AI Challenge is here, focused on trusted software agents. 10 university teams will advance agentic AI for software eng, balancing capability & safety - some will build defenses, others will probe for weaknesses. Apps open Nov 10! amazon.science/nova-ai-challe…

You can learn more here: aws.amazon.com/blogs/aws/amaz…

Today's embeddings systems often require separate models for searching & retrieving docs, text, images, videos, and audio content. With Amazon Nova Multimodal Embeddings, we've built the industry’s first unified model that processes these content types together through a single system. We evaluated the model on a broad range of benchmarks, and it delivers leading accuracy out of the box. But the real impact will be in unlocking insights from unstructured data that was previously difficult to access. Whether it's semantic search across mixed media libraries, building more sophisticated RAG applications, or creating entirely new search experiences, this will make AI more useful for real-world applications!