Nova Act by Amazon

https://nova.amazon.com/act

Overview

The AI landscape has received a powerful new contender. Amazon launched Amazon Nova, a new generation of foundation models designed to deliver frontier intelligence at industry-leading price performance. First announced at AWS re:Invent on December 3, 2024, and expanded with Nova 2 models at AWS re:Invent 2025 on December 3, 2025, Nova promises to generate sophisticated text, code, images, videos, and speech from natural language prompts. The family aims to make top-tier AI more accessible and affordable for developers and enterprises alike, with pricing approximately 75% lower than leading competitors while maintaining competitive performance.

Amazon Nova Act, a browser automation agent service, launched on Product Hunt on December 3, 2025, receiving 139 upvotes and 44 comments. The service achieved general availability status on December 2, 2025, following an earlier research preview release.

Key Features

Amazon Nova is built on a foundation of powerful and practical features across multiple model families. Here’s what makes it stand out:

• Comprehensive Model Family: Nova encompasses multiple specialized models: Nova Micro (text-only), Nova Lite (multimodal – text, images, videos), Nova Pro (multimodal with highest capabilities), Nova Premier (most capable for complex reasoning, launched April 29, 2025), Nova Canvas (image generation), Nova Reel (video generation), Nova Sonic (speech-to-speech), and Nova 2 family including Lite, Pro, Sonic, and Omni released December 2025.
• Frontier-Class Intelligence: Nova models are engineered to compete with advanced models like GPT-4o, Claude 3.5 Sonnet, and Gemini, demonstrating competitive or superior performance on 17 out of 20 benchmarks in independent testing while maintaining significantly lower costs.
• Multimodal Capabilities: Nova models seamlessly process and generate content across different formats, including text, images, videos, code, and speech (depending on model variant), making them versatile tools for diverse applications.
• Industry-Leading Price-Performance Ratio: Nova Micro costs \$0.035 per million input tokens and \$0.14 per million output tokens. Nova Lite costs \$0.06/\$0.24 per million tokens. Nova Pro costs \$0.80/\$3.20 per million tokens—approximately 75% cheaper than Claude 3.5 Sonnet (\$3/\$15) and 65% cheaper than GPT-4o for comparable workloads.
• Native AWS Integration: For businesses embedded in the Amazon Web Services ecosystem, Nova offers seamless integration through Amazon Bedrock, the fully managed AI service platform, simplifying workflows and enhancing security.
• Large Context Windows: Nova models support up to 300,000 tokens for text generation models, with Nova Premier offering a 1 million token context window enabling analysis of 750,000 words, 400-page documents, or 90-minute videos in a single prompt.
• Multilingual Support: Nova understanding models support up to 200 languages, enabling global applications and translations.
• Custom Fine-Tuning and Distillation: Amazon Bedrock enables customers to fine-tune Nova models on proprietary data and use larger models as “teachers” to distill knowledge into smaller, more efficient custom models.
• Nova Forge Service: Announced December 2025, Nova Forge enables enterprises to build custom frontier models (called “Novellas”) by injecting proprietary data during mid-training checkpoints for \$100,000 annually, providing unprecedented customization capabilities.
• Nova Act for Browser Automation: A managed AI agent service powered by a custom Nova 2 Lite model trained with reinforcement learning across thousands of simulated web environments, achieving 90% reliability on browser-based UI workflows for tasks like QA automation, CRM updates, and claims submission.

How It Works

Getting started with Amazon Nova is designed to be straightforward through multiple access points. Users can access the models through Amazon Bedrock (the fully managed AI platform), the nova.amazon.com website (launched March 30, 2025), or through the Amazon Nova Act SDK for browser automation agents.

From there, the process is simple: you provide a natural language prompt describing what you need—whether it’s a block of code, a marketing email, a visual asset for a campaign, or a video clip. Nova processes the request and generates high-quality output in moments.

For Nova Act browser automation, developers break down complex workflows into reliable atomic commands (search, checkout, answer questions about screens) using the SDK. The system supports mixing natural language with Python code, enabling tests, breakpoints, assertions, and thread pooling for parallelization. A no-code playground enables rapid prototyping, while a production console provides monitoring for deployed agent fleets.

For custom model development through Nova Forge, enterprises access pre-trained, mid-trained, or post-trained Nova models and inject their proprietary data to create specialized “Novellas” optimized for specific domain tasks.

Use Cases

The combination of power, versatility, and affordability makes Nova suitable for a wide range of applications organized by business function:

• Enterprise Content Generation: Effortlessly create high volumes of marketing copy, internal documentation, performance reviews, roadmap planning, design docs, incident postmortems, and other business-critical content at scale.
• Coding Assistance: Accelerate development cycles by using Nova to write, debug, and optimize code snippets, translate between programming languages, explain complex algorithms, and execute agentic software engineering tasks.
• Scalable Image and Video Generation: Produce unique, high-quality visual assets for marketing, product design, and creative projects on demand through Nova Canvas (images) and Nova Reel (6-second video clips at 720×1280 pixels with style customization and frame control).
• Cost-Sensitive AI Applications: Build and deploy powerful AI-driven features and applications where budget is a key consideration, thanks to Nova’s competitive pricing approximately 75% lower than leading alternatives.
• Document and Video Analysis: Leverage large context windows to summarize charts, digest 400-page documents, analyze 90-minute videos, and extract insights from multimodal content.
• Browser Automation and QA Testing: Use Nova Act to automate form submissions, claims processing, CRM updates, calendar management, and comprehensive quality assurance workflows with 90% reliability, reducing testing cycles from weeks to hours.
• Conversational AI: Deploy Nova 2 Sonic for real-time, multilingual speech-to-speech applications with seamless switching between voice and text modes.
• Unified Multimodal Generation: Utilize Nova 2 Omni to analyze entire product catalogs, testimonials, brand guidelines, and video libraries simultaneously, generating complete marketing campaigns including headlines, copy, social posts, and visuals in one workflow.

Pros \& Cons

No tool is perfect. Here is a balanced look at Amazon Nova’s strengths and potential weaknesses.

Advantages

• Deep AWS Integration: For the millions of developers and businesses on AWS, Nova is a natural fit, offering streamlined integration, unified billing, and security within the Amazon Bedrock fully managed service.
• Extremely Competitive Pricing: Nova delivers approximately 75% cost savings compared to Claude 3.5 Sonnet and 65% cost savings compared to GPT-4o, making state-of-the-art AI more accessible to startups, SMBs, and large enterprises. Independent benchmarking shows savings of \$5,674 per month (\$68,098 annually) for organizations running 10 million queries monthly.
• Comprehensive Model Family: Unlike point solutions, Nova offers specialized models for text generation, multimodal understanding, image generation, video generation, speech-to-speech, and browser automation, providing end-to-end AI capabilities.
• Strong Competitive Performance: Independent benchmarks show Nova Pro competing effectively with GPT-4o and Claude 3.5 Sonnet on 17 out of 20 benchmarks while maintaining superior cost-effectiveness.
• Enterprise Customization: Nova Forge enables unprecedented custom frontier model development with proprietary data integration during training, while fine-tuning and distillation capabilities allow domain-specific optimization.
• High Reliability for Automation: Nova Act achieves 90% reliability on browser-based UI workflows, transforming quality assurance from weeks to hours with proven enterprise deployments.

Disadvantages

• Saturated Market: Nova enters a crowded field and must compete for attention against highly established and trusted players like OpenAI, Google, and Anthropic, which already have significant market share and mindshare.
• Performance Gaps on Specialized Benchmarks: Nova Premier performs below Gemini 2.5 Pro on certain coding tests (SWE-Bench Verified) and shows weaker performance on specialized benchmarks measuring advanced math (GPQA Diamond, AIME 2025) and science knowledge.
• Limited Availability for Newest Models: Nova Premier launched several months after initial announcement (April 29, 2025), and Nova 2 Omni is available only through early access via Nova Forge customers, creating adoption delays.
• Pricing Complexity: While overall costs are lower, the multiple model variants with different pricing tiers (Micro, Lite, Pro, Premier, Canvas, Reel, Sonic, Omni) create complexity for organizations determining optimal model selection.
• AWS Ecosystem Dependency: Organizations not already using AWS may face migration complexity, while the tight AWS integration creates potential vendor lock-in concerns.

How Does It Compare?

Amazon Nova vs. OpenAI GPT-4o

OpenAI GPT-4o is a flagship multimodal model from OpenAI known for advanced reasoning and creative capabilities.

Pricing Comparison:

• Amazon Nova Pro: \$0.80 input / \$3.20 output per million tokens
• GPT-4o: Approximately \$2.50 input / \$10 output per million tokens (varies by tier)
• Cost Savings: Nova Pro delivers 65-68% cost savings compared to GPT-4o for comparable workloads

Performance:

• Benchmarks: Independent testing shows Nova Pro competitive or equal to GPT-4o on 8 out of 16 benchmarks with superior price-performance
• Strengths: GPT-4o excels at creative writing, complex reasoning, and established brand trust; Nova Pro excels at cost-effectiveness and AWS integration

Context Window:

• Nova Pro: 300,000 tokens
• Nova Premier: 1 million tokens
• GPT-4o: 128,000 tokens

When to Choose Amazon Nova: For AWS-native environments, cost-sensitive applications requiring high-volume inference, and when 75% cost savings justify slight performance trade-offs.
When to Choose GPT-4o: For maximum creative capabilities, established enterprise trust, non-AWS environments, and when cutting-edge performance outweighs cost considerations.

Amazon Nova vs. Anthropic Claude 3.5 Sonnet

Anthropic Claude 3.5 Sonnet is a highly capable model known for strong reasoning, coding abilities, and safety alignment.

Pricing Comparison:

• Amazon Nova Pro: \$0.80 input / \$3.20 output per million tokens
• Claude 3.5 Sonnet: \$3 input / \$15 output per million tokens
• Cost Savings: Nova Pro is 4.7× cheaper (approximately 75% cost reduction)

Performance:

• Nova Pro performs comparably to Claude 3.5 Sonnet on most benchmarks while maintaining dramatic cost advantages
• Claude 3.5 Sonnet demonstrates superior performance on specialized coding tasks and complex reasoning

Context Window:

• Both: Support 200,000+ token contexts

When to Choose Amazon Nova: For cost-sensitive AWS deployments, high-volume applications, and when Nova Pro’s competitive performance meets requirements at 4.7× lower cost.
When to Choose Claude 3.5 Sonnet: For maximum reasoning quality, specialized coding tasks, and when budget permits premium pricing for incremental performance gains.

Amazon Nova vs. Google Gemini 1.5 / 2.5

Google Gemini is Google’s multimodal AI model family with strong multimodal understanding and competitive pricing.

Pricing Comparison:

• Amazon Nova Pro: \$0.80 / \$3.20 per million tokens
• Gemini 1.5 Pro: \$1.25 / \$5.00 per million tokens
• Gemini 2.5 Pro: Similar tier to Gemini 1.5 Pro
• Cost Advantage: Nova Pro provides significant savings

Performance:

• Nova 2 Lite: Competitive with Gemini Flash 2.5 with stronger performance on document processing and video understanding
• Nova 2 Pro: Competitive with Gemini 2.5 Pro with strengths in multi-document analysis and agentic workflows
• Gemini Advantages: Superior performance on SWE-Bench Verified coding tasks and certain specialized benchmarks

When to Choose Amazon Nova: For AWS ecosystems, cost-sensitive deployments, and when competitive performance at lower cost is priority.
When to Choose Google Gemini: For Google Cloud environments, maximum multimodal capabilities, and when Google ecosystem integration is critical.

Amazon Nova vs. Open-Source Models (Llama, Mistral)

Open-source models like Meta’s Llama and Mistral AI offer self-hosted alternatives.

Cost Structure:

• Amazon Nova: Pay-per-token cloud service through AWS Bedrock
• Open-source: Infrastructure costs for self-hosting (compute, storage, maintenance)

Ease of Use:

• Amazon Nova: Fully managed service with no infrastructure management
• Open-source: Requires technical expertise for deployment, scaling, and maintenance

Customization:

• Amazon Nova: Fine-tuning through Bedrock, Nova Forge for custom frontier models (\$100K/year)
• Open-source: Complete control over model architecture and training

When to Choose Amazon Nova: For rapid deployment, managed services, enterprise support, and when development speed outweighs infrastructure control.
When to Choose Open-Source: For maximum customization, avoiding cloud dependencies, and when in-house ML infrastructure expertise exists.

Amazon Nova Act vs. Robotic Process Automation (UiPath, Automation Anywhere)

Traditional RPA tools automate repetitive tasks through UI interaction recording.

Automation Approach:

• Nova Act: AI-native agents using reinforcement learning with 90% reliability on complex workflows
• Traditional RPA: Rule-based automation requiring explicit programming of each step

Flexibility:

• Nova Act: Adapts to UI changes and handles variations in web environments
• Traditional RPA: Brittle; breaks when UI elements change

Development Time:

• Nova Act: Minutes to prototype in no-code playground; rapid iteration
• Traditional RPA: Weeks to develop, test, and maintain automation scripts

Pricing:

• Nova Act: AWS service pricing (pay-per-use model)
• Traditional RPA: Enterprise licensing with bot costs, infrastructure, and dedicated RPA teams

When to Choose Nova Act: For modern SaaS applications, rapid deployment, AI-adaptive workflows, and browser-based automation.
When to Choose Traditional RPA: For legacy desktop applications, highly regulated environments with strict audit requirements, and when existing RPA infrastructure exists.

Final Thoughts

Amazon Nova represents a strategic and powerful entry into the foundation model race. It’s not just another AI model; it’s an ecosystem play designed to democratize access to frontier AI through aggressive pricing and seamless AWS integration.

The December 2024 launch of Nova’s initial family (Micro, Lite, Pro, Premier, Canvas, Reel) followed by the December 2025 Nova 2 expansion (Lite, Pro, Sonic, Omni) demonstrates Amazon’s commitment to rapid innovation. The introduction of Nova Forge for custom frontier model development and Nova Act for reliable browser automation positions Amazon uniquely in addressing enterprise customization and agentic automation needs.

Independent benchmarking validates Nova’s value proposition: approximately 75% cost savings compared to Claude 3.5 Sonnet and 65% cost savings compared to GPT-4o, while maintaining competitive performance on 17 out of 20 benchmarks. Organizations running 10 million queries monthly can save \$68,000 annually by switching to Nova Pro from GPT-4o.

However, prospective users should consider: Nova faces fierce competition from established players with strong brand recognition; performance gaps exist on specialized benchmarks requiring maximum capabilities; and AWS ecosystem integration, while powerful for existing customers, creates potential vendor lock-in for new adopters.

For organizations already embedded in AWS, Nova presents a compelling value proposition through unified billing, native integration with Amazon Bedrock, and dramatic cost savings. For cost-sensitive applications requiring high-volume AI inference—such as customer support automation, content generation at scale, and document analysis—Nova’s price-performance ratio is industry-leading.

The proven 90% reliability of Nova Act for browser automation represents a breakthrough for quality assurance teams, with early customers like Hertz reporting 5× acceleration in shipping velocity and Leo Commerce reducing comprehensive test creation from weeks to minutes.

While Nova faces challenges competing against established leaders, its unique value proposition—frontier intelligence at 75% lower cost with deep AWS integration—makes it a tool that anyone serious about implementing cost-effective, scalable AI should evaluate closely. For AWS customers, Nova is increasingly difficult to ignore; for others, the cost savings may justify AWS adoption.

https://nova.amazon.com/act