AWS-native AI applications
Build generative features where data, identity, networking, monitoring, and production infrastructure already live in AWS.
Independent tool overview
Amazon Nova at a glance
Amazon Nova is AWS's family of foundation models and services for multimodal understanding, reasoning, voice, creative generation, embeddings, browser agents, and deeply customized enterprise models.
Visit the official Amazon Nova site ↗
- Provider
- Amazon Web Services
- Product type
- Foundation-model and agent-service portfolio
- Primary access
- Amazon Bedrock APIs, Nova developer tools, and AWS consoles
- Core modalities
- Text, images, video, documents, speech, and embeddings
- Current generation
- Nova 2, alongside supported Nova 1 creative and understanding models
- Long context
- Up to 1 million tokens on documented Nova 2 models
- Agent pricing
- Nova Act is listed at $4.75 per agent-hour
- Last reviewed
- August 30, 2026
Overview
What Amazon Nova is
Amazon Nova is not a single consumer chatbot. It is an AWS portfolio for developers and enterprises building AI applications through Amazon Bedrock and related Nova services. The portfolio spans cost-oriented multimodal models, real-time speech, image and video generation, embeddings, browser automation through Nova Act, and custom frontier-model development through Nova Forge.
Nova is most compelling for teams already operating on AWS that want model, agent, data, security, and deployment controls in one cloud. The breadth can also make evaluation harder: model availability, modalities, context limits, regions, service tiers, customization paths, and pricing differ across the portfolio. Treat Amazon's benchmark and price-performance claims as provider-reported, then test Nova against your own tasks, latency targets, safety cases, and total AWS bill.
Use cases
Who Amazon Nova is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
High-volume multimodal processing
Use Nova 2 Lite for cost-sensitive document, image, video, coding, extraction, and agentic workloads.
Real-time voice agents
Use Nova 2 Sonic for speech-to-speech experiences that mix conversation, text, tools, and long-running context.
Browser workflow automation
Use Nova Act to build and manage agents that carry out bounded tasks through web interfaces.
Deep enterprise customization
Evaluate Nova Forge when ordinary prompting, RAG, and fine-tuning are insufficient and proprietary data must shape a dedicated model.
Cross-modal search and RAG
Use Nova Multimodal Embeddings to index and retrieve related information across text, images, documents, video, and audio.
Capabilities
Core Amazon Nova features
Nova 2 Lite
Cost-oriented reasoning model for text, image, video, and document inputs with text output and optional extended thinking.
Nova 2 Pro
Higher-capability multimodal reasoning model announced for complex planning, coding, document analysis, video reasoning, and knowledge distillation; confirm current access path and region.
Nova 2 Sonic
Streaming speech model for real-time voice interactions that can combine spoken and text input and output.
Nova 2 Omni
Announced unified model that can process text, images, video, and speech and generate text and images; availability has included Nova Forge early access.
Multimodal embeddings
Create a shared embedding representation for semantic search, recommendations, similarity, and agentic RAG across several media types.
Nova Canvas and Reel
Generate or transform images with Canvas and create video with Reel for creative and marketing workflows.
Built-in tools
Supported Nova 2 models can use AWS-provided web grounding and code interpretation without a separately built tool integration.
Long context
AWS documents up to one million tokens of context for Nova 2 models, supporting large documents, codebases, videos, and extended conversations.
Bedrock inference
Invoke models through Bedrock Converse or model-specific Invoke APIs with streaming, tool use, guardrails, and multimodal inputs.
Model customization
Use supported supervised fine-tuning, preference optimization, reinforcement learning, distillation, or continued training paths depending on model and service.
Nova Forge
Annual-subscription service for building dedicated Nova variants, called Novellas, using proprietary data and access to training checkpoints and tools.
Nova Act
Develop, test, deploy, monitor, and manage browser-based agents using a Nova model trained for UI workflows.
AWS production controls
Combine Nova with Bedrock features, IAM, networking, observability, knowledge bases, guardrails, and region-specific infrastructure.
Process
How the Amazon Nova workflow works
Step 1
Define one measurable workload
Specify the input modalities, output contract, acceptable error rate, latency, throughput, regions, privacy needs, and maximum unit cost.
Step 2
Choose the smallest viable Nova path
Start with Nova 2 Lite or another available base model before adding reasoning, larger models, Nova Act, fine-tuning, or Forge.
Step 3
Build a representative evaluation set
Include ordinary cases, rare edge cases, adversarial inputs, sensitive data, tool failures, long-context examples, and known correct answers.
Step 4
Prototype through Bedrock
Use the recommended API, lock the model ID and region, define structured outputs, set timeouts, and log requests, responses, latency, and token usage safely.
Step 5
Ground and constrain
Add RAG, tool schemas, guardrails, permissions, citations, and human approval where the model could make claims or take consequential actions.
Step 6
Run cost and reliability tests
Measure full workload cost—including reasoning, tools, retries, media, storage, network, agent time, and monitoring—rather than token price alone.
Step 7
Deploy with continuous evaluation
Version prompts and models, monitor drift and failures, maintain fallbacks, sample production outputs, and requalify the system before changing models or regions.
Cost
Amazon Nova pricing and free plan
Amazon Nova uses several pricing models. Foundation models are metered by tokens, media, embeddings, or generation units through Bedrock; Nova Act is billed by agent-hour; and Nova Forge requires an annual subscription. Rates vary by model, region, service tier, input type, and customization path.
Nova foundation models
Usage-based
Pay for model inference based on the selected Nova model and the amount and type of input and output processed.
- Understanding models generally meter input and output tokens
- Speech, image, video, and embedding workloads use model-specific units
- Standard, Priority, and Flex service tiers can trade cost against allocation and responsiveness
- Cross-region inference and region availability vary
- Use the live Nova and Bedrock pricing tables for workload estimates
Nova Act
$4.75 per agent-hour
Browser-workflow agents are billed for real-world elapsed time while each agent is actively working.
- Parallel agents each generate their own agent-hours
- AWS says human-in-the-loop waiting time is excluded
- Retries, testing, orchestration, storage, and other AWS services can add cost
Nova Forge
Annual subscription
Enterprise model-development service with pricing available through the Nova Forge console or AWS sales.
- Subscription provides access to the Forge service
- Training, data preparation, evaluation, deployment, and inference costs should be modeled separately
- Designed for organizations with substantial proprietary data and model-development resources
Custom Nova inference
Usage-based
AWS states on-demand inference for supported custom Nova models is priced at the corresponding base-model inference rate.
- Customization jobs and supporting SageMaker or Bedrock resources add separate charges
- Provisioned throughput can have a different cost structure
- Confirm model, training date, deployment type, and region eligibility
Pricing checked . Check current pricing at the source ↗
Assessment
Amazon Nova strengths and limitations
Where it stands out
- Broad model family spans understanding, reasoning, speech, creative generation, embeddings, and agents
- Strong fit for organizations already standardized on AWS
- Bedrock provides a managed API and integration point for production applications
- Nova 2 Lite targets high-volume cost-sensitive multimodal work
- One-million-token context supports unusually large inputs on documented Nova 2 models
- Built-in web grounding and code interpretation reduce some integration work
- Nova Act provides a managed route from browser-agent prototype to deployment and monitoring
- Nova Forge offers a deeper customization path than prompt tuning or ordinary fine-tuning
- Several inference service tiers let teams trade cost, latency, and compute priority
- Multimodal embeddings simplify retrieval across mixed media
What to consider
- Amazon Nova is a developer platform, not a ready-to-use consumer assistant
- Production use requires AWS architecture, IAM, monitoring, evaluation, and cost-management expertise
- Nova models are primarily accessed through Amazon Bedrock, creating meaningful AWS platform dependence
- The portfolio has many models and services with different regions, previews, modalities, limits, and billing units
- Nova 2 Pro and Omni availability may differ from their announcement positioning and should be checked before design decisions
- Provider-reported benchmark and price-performance claims may not predict results on a specific workload
- Long context does not guarantee accurate retrieval, reasoning, citation, or instruction following across the entire input
- Reasoning, web grounding, code execution, tools, media, and retries can materially increase latency and cost
- Model outputs can hallucinate, omit important context, reproduce bias, or fail inconsistently
- Browser agents can click the wrong control, submit incorrect data, expose credentials, or be manipulated by untrusted page content
- Nova Act's published reliability figures are based on provider-selected early customer workflows, not a universal guarantee
- Nova Forge demands high-quality proprietary data, ML expertise, governance, and a business case large enough to justify an annual service
- Cross-region inference can conflict with strict data-residency or processing-location requirements
- Creative image and video output still requires rights, consent, brand, and safety review
- Fine-tuning and custom models introduce versioning, evaluation, rollback, and lifecycle obligations
- AWS service fees beyond the model—such as S3, networking, observability, vector storage, and orchestration—affect total cost
Compare
Amazon Nova alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Consumer
Gemini 3.1 Pro
Choose Gemini 3.1 Pro when Google Cloud and Google's multimodal model ecosystem are a better infrastructure fit.
Explore Gemini 3.1 Pro →Consumer
Claude Opus 4.6
Choose Claude Opus 4.6 when Anthropic's model behavior and agentic capabilities outperform Nova on your evaluated workload.
Explore Claude Opus 4.6 →Consumer
GPT 5.4
Choose GPT-5.4 when OpenAI's API ecosystem, computer use, and model performance better match the application requirements.
Explore GPT 5.4 →Questions
Amazon Nova FAQs
What is Amazon Nova?
Amazon Nova is AWS's portfolio of foundation models and services for multimodal understanding, reasoning, speech, creative generation, embeddings, browser automation, and custom enterprise models.
Is Amazon Nova a chatbot?
No. It is primarily a developer and enterprise platform. Teams call Nova models through Amazon Bedrock or use services such as Nova Act and Nova Forge to build applications.
What is the difference between Nova 2 Lite and Nova 2 Pro?
Nova 2 Lite is positioned for fast, cost-efficient high-volume reasoning and multimodal work. Nova 2 Pro is positioned for harder planning, coding, document, video, and agentic tasks, but teams should confirm its current access and region availability.
What is Nova Act?
Nova Act is AWS's service for building and operating agents that complete tasks through browser interfaces. AWS currently lists it at $4.75 per active agent-hour.
What is Nova Forge?
Nova Forge is an annual-subscription service that lets qualified organizations build dedicated Nova variants using proprietary data and access to Nova training checkpoints, data-mixing, reinforcement-learning, distillation, and safety tools.
How much does Amazon Nova cost?
Foundation-model pricing is usage-based and varies by model, modality, region, and service tier. Nova Act is listed at $4.75 per agent-hour, while Nova Forge uses annual subscription pricing available through AWS.
Does Amazon Nova support long context?
AWS documents up to one million tokens of context for current Nova 2 models. Teams should still test retrieval and reasoning accuracy at the actual input length they plan to use.
Can Amazon Nova process video and audio?
Yes, depending on the model. Nova includes models for video and document understanding, speech interaction, image and video generation, and multimodal embeddings.
How should a team evaluate Amazon Nova?
Run a representative task set against the smallest suitable Nova model and alternatives, then compare accuracy, structured-output reliability, safety, latency, regional availability, integration work, and total cost including tools and retries.
Bottom line
Our Amazon Nova verdict
Amazon Nova is a serious option for AWS-native teams that want a wide multimodal model portfolio, long context, managed Bedrock access, browser agents, and an unusually deep customization path. Its value is architectural rather than conversational: Nova can consolidate several AI workloads inside AWS. The buying decision should come from workload-specific evaluation, because the portfolio's breadth, changing availability, provider-reported benchmarks, and multi-part pricing make headline comparisons insufficient.
Visit Amazon Nova website ↗Similar Consumer Tools
View all Consumer AI tools →
Mistral 3
Mistral 3 - Mistral's next generation of open-source models including Small, Large, and Ministral variants.
Kling AI Avatar 2.0
Kling Avatar 2.0 - Upgraded avatar model with enhanced expressiveness, realism, and full 5-minute generations
Shopping Research
Shopping Research - Personalized, interactive shopping experience directly in ChatGPT to find the perfect products
Purpose
Purpose - AI mentor for deep, personalized guidance on-demand
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.