Amazon Nova Pricing & API Models
Compare Amazon Nova pricing by input tokens, output tokens, context length, and capability. The model rows below come from the live AI Pricing Hub database, so use the official AWS links to verify the exact region, tier, modality, and account terms before launch.
5 Nova model rows are currently available. Last stored price update: Aug 8, 2026.
Amazon Nova pricing at a glance
- Input and output tokens are shown separately.
- Nova models are compared in USD per 1M tokens.
- Use the calculator for a workload-specific estimate.
Amazon Nova models and token pricing
Use the table to shortlist models, then open the calculator with the model preselected. Prices are normalized to USD per 1M input or output tokens where the tracked provider row supplies a token rate.
| Model | Input / 1M | Output / 1M | Context | Capabilities | Cost |
|---|---|---|---|---|---|
| Nova 2 Lite Amazon Bedrock |
$0.30 | $2.50 | 1.0M | chatreasoningtool_usevideovision |
|
| Nova Premier 1.0 Amazon Bedrock |
$2.50 | $12.50 | 1.0M | chattool_usevision |
|
| Nova Micro 1.0 Amazon Bedrock |
$0.04 | $0.14 | 128k | chattool_use |
|
| Nova Lite 1.0 Amazon Bedrock |
$0.06 | $0.24 | 300k | chattool_usevision |
|
| Nova Pro 1.0 Amazon Bedrock |
$0.80 | $3.20 | 300k | chattool_usevision |
The displayed row is a planning reference, not an AWS invoice. Region, service tier, modality, caching, batch processing, priority, and additional Bedrock features can change the final charge.
Which Amazon Nova model should you choose?
Nova Micro
Start with Micro when latency and high-volume text work matter more than multimodal input. It is a practical fit for classification, short summaries, routing, and simple support replies.
Nova Lite
Choose Lite for lower-cost multimodal workloads such as image understanding, document extraction, and customer-support flows that need more than text but do not require a flagship model.
Nova Pro
Use Pro when image or document reasoning quality is more important than the lowest token price. Compare output cost carefully if responses contain long explanations or structured results.
Nova Premier and Sonic
Premier is aimed at more demanding reasoning and distillation workflows, while Sonic is relevant to speech and real-time voice scenarios. Treat these as separate workload decisions rather than assuming one family-wide price.
How Amazon Nova pricing works
For a token-priced workload, a useful first estimate is input tokens × input price + output tokens × output price. If a request uses 200,000 input tokens and 20,000 output tokens, calculate each side separately instead of applying one blended rate. This matters because code, reasoning, and document workflows often produce much longer outputs than short chat replies.
Keep the model row and the billing context together. The same Nova family can have different rates or conditions depending on the Amazon Bedrock region, standard or priority service tier, batch or asynchronous processing, modality, caching, and account-level terms. Use the live table for a shortlist, then verify the exact AWS price before procurement or production rollout.
For a monthly budget, multiply the per-request estimate by request volume and add a buffer for retries, tool calls, safety checks, long-context spikes, and traffic that falls back to a different model.
A practical cost checklist
- Choose the Nova model and region.
- Measure representative input and output tokens.
- Check whether batch, cache, priority, or voice charges apply.
- Estimate retries and monthly request volume.
- Verify the result against AWS pricing before launch.
Amazon Nova cost examples by workload
These examples show how to frame the calculation. They are usage assumptions, not an invoice quote, and the live model table supplies the rate used in the final estimate.
| Workload | Example monthly usage | What usually drives cost | Good first check |
|---|---|---|---|
| Support assistant | 500k input + 100k output tokens | Request volume and response length | Compare Micro and Lite total cost |
| Document extraction | 5M input + 1M output tokens with images or pages | Multimodal input, retries, and context size | Test Lite before Pro |
| Reasoning workflow | 1M input + 300k output tokens | Long output, retries, and review rate | Compare Pro/Premier quality against output price |
| Voice or real-time flow | Token usage plus audio duration | Audio units, latency tier, and session volume | Separate token cost from voice-specific charges |
A fair comparison uses the same prompt set, output cap, request volume, region, and service assumptions for every candidate. If a model reduces retries or manual review, include that operational effect in the final decision instead of comparing token rates in isolation.
What the Nova pricing table does not include
The table is intentionally focused on model and token pricing. It does not promise a complete AWS invoice because a production request can include charges outside the base input/output row. Teams may need to account for image or audio units, batch or priority processing, provisioned capacity, regional availability, routing, storage, retrieval, orchestration, and other AWS services around the model call.
This boundary is useful: it keeps model selection comparable while making the assumptions visible. Open the Bedrock calculator when your decision depends on service-level conditions, and keep separate line items for infrastructure or application costs that are not part of the model token rate.
Before you commit production traffic
- Pin the model identifier and confirm that it is available in the intended AWS region.
- Record input, output, cached, batch, image, audio, and tool-related assumptions separately.
- Run a representative quality and latency test, not only a price sort.
- Add a retry and fallback budget for malformed responses, rate limits, and provider routing.
- Recheck the official AWS source when the model, tier, region, or workload changes.
Amazon Nova vs Amazon Bedrock pricing
Nova is a model family; Bedrock is the managed AWS access layer. Keeping those two concepts separate prevents a common pricing comparison error.
| What you are comparing | What to check | Useful next step |
|---|---|---|
| Amazon Nova model | Input/output token rate, context, modality, quality, and model availability | Review the Nova rows |
| Amazon Bedrock access | Region, service tier, batch or priority mode, routing, and feature charges | Open the Bedrock calculator |
| Monthly workload | Tokens per request, requests per month, retries, and fallback traffic | Estimate monthly cost |
For authoritative rates and availability, verify the current AWS Amazon Nova pricing page and the Amazon Nova documentation.
Amazon Nova pricing FAQ
What is the cheapest Amazon Nova model?
The cheapest choice depends on the current model rows and your input/output mix. Start by sorting the live table by input and output price separately, then confirm that the model supports the modality and context your workload needs.
Is Amazon Nova pricing the same as Amazon Bedrock pricing?
No. Nova names the model family, while Bedrock is the AWS service used to access models. Your final estimate can also depend on region, service tier, modality, batch or priority processing, and account terms.
How do I compare Nova Lite and Nova Pro cost?
Compare both input and output prices using the same token volume, then test representative prompts. Lite may be a better budget fit for multimodal extraction, while Pro can be worth the premium when reasoning quality reduces retries or manual review.
Where should I verify Amazon Nova pricing?
Use this page for a normalized model comparison and workload estimate, then verify the exact model, region, tier, and feature charges on the official AWS pricing and documentation pages before deployment.
Continue your Amazon Nova cost comparison
Use the focused model page for Nova, the Bedrock calculator for service-level assumptions, or the comparison tools for alternatives.