Amazon API Pricing, Models & Rate Limits (2026)
Amazon trains the Nova family and serves it exclusively through AWS Bedrock's Converse API — no standalone Nova endpoint exists outside AWS. That means IAM/SigV4 auth and AWS billing instead of a bearer API key, which is the single biggest adoption-cost difference on this page.
Also known as: AWS Bedrock, Nova.
How much does the Amazon API cost?
Amazon API pricing is usage-based and varies by model and token volume. Use the current model table and provider documentation to estimate a workload before committing to production.
For cross-provider quota units and fixed-workload capacity, see the LLM API rate-limit comparison; this page remains the authoritative owner for Amazon provider facts.
Amazon product vs API
Amazon consumer access and API billing are separate surfaces; check the provider documentation for current account terms.
Three decisions unique to Amazon
Amazon current-model price mechanics
| Current model | Input | Cached input | Output | Batch | Verified |
|---|
| Amazon Nova Micro | $0.035/M | Unavailable — no model cache rate | $0.140/M | 50% off eligible Batch API | 2026-06-14 |
|---|
| Amazon Nova Lite | $0.060/M | Unavailable — no model cache rate | $0.240/M | 50% off eligible Batch API | 2026-06-14 |
|---|
| Amazon Nova Pro | $0.800/M | Unavailable — no model cache rate | $3.200/M | 50% off eligible Batch API | 2026-06-14 |
|---|
Messages API vs OpenAI compatibility map
| Choice | Decision rule | Evidence |
|---|
| Billing | ChatGPT plan never includes API credits | Separate metered API account |
|---|
| Input/cached/output | Token prices are model rows | Use calculator for workload totals |
|---|
| Limits/auth | Per-account, per-region AWS service quotas · Bearer API key | Verify before production |
|---|
Adoption map: what is documented versus unavailable
| Dimension | Recorded value | Decision consequence |
|---|
Try Amazon side by side →Verified 2026-08-14. dated provider pricing/source →
Price range /M
$0.06–$1.40
Amazon model pricing
Compare Amazon models by input, output, and blended token cost below.
Pricing values are registry-backed and were most recently verified on 2026-06-14. Sources: https://aws.amazon.com/bedrock/pricing. Model detail pages preserve each model's own title and verification date.
* Blended comparison assumes 3 input tokens for every output token; it is not the provider's billing unit.
Speed
Best for
Build with Amazon
Amazon implementation details
Verified 2026-08-14 against source.
Amazon publishes its current authentication, limits, and data-handling details in the linked documentation.
| OpenAI-compatible | No |
| API base URL | bedrock-runtime.{region}.amazonaws.com (Converse API) |
| Auth model | AWS SigV4 (IAM credentials) |
| Prompt caching | Not documented |
| Batch discount | 50% |
| Free tier | No free tier |
| Free-tier limits | No provider-wide Nova API free tier is published; AWS account promotions, if any, are separate. |
| Free-tier expiry | Not published |
| Rate-limit model | Per-account, per-region AWS service quotas |
| Data residency | Selectable AWS region |
| Trains on API data | No |
| SLA published | Yes |
Switching to and from Amazon
Full Amazon alternatives comparison →Calling Amazon through All AI Ask
Calling Amazon directly means signing every request with AWS SigV4 and managing IAM credentials instead of a single bearer key — our gateway removes that: every model, including Amazon's, is called the same way.
curl https://allaiask.com/api/v1/chat \
-H "Authorization: Bearer $ALLAIASK_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "nova-micro", "messages": [{"role": "user", "content": "Hello"}]}'FAQ
Is Amazon OpenAI-compatible?
No. Amazon uses its own request format and auth model (AWS SigV4 (IAM credentials)), so OpenAI client code needs to be adapted.
Does Amazon support prompt caching?
Not documented as of 2026-08-14 — we did not find a published prompt-caching feature for Amazon. If that changes, this page updates.
Does Amazon have a free tier?
No free tier is published as of 2026-08-14. No provider-wide Nova API free tier is published; AWS account promotions, if any, are separate.
How much does the Amazon API cost?
Current Amazon models range from $0.06 to $1.40 per million blended tokens (3:1 input:output). Full per-model pricing is in the table below.
Where is Amazon API data hosted?
Selectable AWS region
Batch 48 · amazon provider adoption evidence. Surface verification: 2026-08-14. Missing joins are deliberately published as Unavailable.
Bedrock inference-surface decision compiler
Frozen Batch 48 amazon fixture: Stateless chat, stateful agent, Anthropic-native, cross-model, direct-control, and unsupported fixtures. First-party evidence checked 2026-08-14; unresolved joins fail closed. Formula / decision rule: surface pass = endpoint API ∧ model support ∧ state owner ∧ auth ∧ stream shape
| Frozen fixture | Inputs and observation | Formula / boundary | State |
|---|
batch48-amazon-m1-r1 stateless chat · Converse | endpoint=bedrock-runtime; API=Converse; model=amazon.nova-pro-v1:0; auth=SigV4; state=caller; stream=ConverseStream Converse owns the request state and the model is available through the regional Bedrock runtime. | eligible = API ∧ region model access ∧ IAM action ∧ stream contract A model ID without region access is not an eligible route. | PASS — regional Bedrock surface. |
batch48-amazon-m1-r2 stateful agent · Anthropic-native workload | endpoint=Agents for Bedrock; API=InvokeAgent; state=agent alias; model=provider-managed; auth=IAM; feature=agent action Agent state and action custody belong to Bedrock Agents; native provider semantics are not implied. | pass = control-plane agent ID ∧ runtime alias ∧ IAM action ∧ action group Converse success cannot prove agent readiness. | UNAVAILABLE — agent control-plane join missing. |
batch48-amazon-m1-r3 cross-model app · direct model-specific control · unsupported pair | app=multi-model; model=Nova; API=Converse; control=provider-native; endpoint=unsupported; correction=use model adapter The app can select a model through Converse, but the native control has no joined Bedrock field. | control pass = submitted control ∧ Bedrock-supported field ∧ effective echo Unsupported controls must not be serialized as successful. | FAIL CLOSED — adapter boundary retained. |
Provenance: Frozen Batch 48 amazon fixture: module 1 observation and calculation. First-party evidence checked 2026-08-14; unresolved joins fail closed. Surface verification date: 2026-08-14. AWS Bedrock User Guide (first-party source)
Region–IAM–model-access handshake matrix
Frozen Batch 48 amazon fixture: Account, principal, action/resource, region, network, encryption, and quota joins. First-party evidence checked 2026-08-14; unresolved joins fail closed. Formula / decision rule: handshake = principal ∧ action/resource ∧ region ∧ model entitlement ∧ network ∧ logs
| Frozen fixture | Inputs and observation | Formula / boundary | State |
|---|
batch48-amazon-m2-r1 same account · same region | account=111; role=bedrock-prod; region=us-east-1; action=bedrock:InvokeModel; resource=arn:model/nova; VPC=vpce-17; quota=account-region Principal, resource, region, and private network join; the quota owner is the AWS account in that region. | effective gate = IAM ∧ region access ∧ VPC route ∧ account quota An IAM pass in us-east-1 cannot be moved to eu-west-1. | PASS — all gates joined. |
batch48-amazon-m2-r2 cross-account role · KMS-controlled logs | caller=222; role=arn:aws:iam::111:role/cross; region=us-west-2; KMS=key-88; logs=group-9; model access=Unavailable Trust and log encryption join, but target-account model access is unverified. | eligible = trust policy ∧ action/resource ∧ target model access ∧ log key A valid role assumption is not model entitlement. | UNAVAILABLE — target model gate missing. |
batch48-amazon-m2-r3 VPC-only · cross-region inference profile · unavailable model-region | network=private; profile=us.anthropic.claude; requested region=ap-south-1; model region=Unavailable; failure gate=region The private network is valid, but no joined model-region route exists for the requested profile. | route = profile region mapping ∧ model access ∧ network path Cross-region profiles do not authorize arbitrary regional residency claims. | FAIL CLOSED — select an evidenced region. |
Provenance: Frozen Batch 48 amazon fixture: module 2 observation and calculation. First-party evidence checked 2026-08-14; unresolved joins fail closed. Surface verification date: 2026-08-14. AWS Bedrock User Guide (first-party source)
Managed-capability dependency ledger
Frozen Batch 48 amazon fixture: Guardrails, knowledge bases, agents, batch, evaluations, prompts, and imported models. First-party evidence checked 2026-08-14; unresolved joins fail closed. Formula / decision rule: managed-ready = control-plane ID ∧ runtime ID ∧ region ∧ custody ∧ lifecycle ∧ rollback
| Frozen fixture | Inputs and observation | Formula / boundary | State |
|---|
batch48-amazon-m3-r1 guardrail · knowledge retrieval | guardrail=gr-4; knowledgeBase=kb-2; region=us-east-1; runtime=Converse; custody=AWS; rollback=platform Both managed IDs join the runtime region; rollback remains a Bedrock control-plane action. | ready = each feature ID joins same region and runtime Successful model invocation cannot prove knowledge-base readiness. | PASS WITH OWNER — managed features remain AWS-owned. |
batch48-amazon-m3-r2 agent action · batch invocation · evaluation job | agent=ag-7; actionGroup=agx-1; batch=batch-3; eval=eval-9; runtime IDs=partial; lifecycle=joined Agent and lifecycle IDs join, but batch runtime settlement is absent. | coverage = joined managed IDs / declared IDs = 3/4 = 75% No 100% coverage claim from partial managed state. | UNAVAILABLE — batch settlement missing. |
batch48-amazon-m3-r3 prompt management · imported model workflow | prompt=pm-11; imported model=custom-2; region=Unavailable; custody=AWS; blast radius=workflow; rollback=Unavailable Prompt identity joins, while imported-model region and rollback owners do not. | rollback-safe = region ∧ custody ∧ lifecycle ∧ rollback owner Prompt success cannot establish imported-model portability. | FAIL CLOSED — retain source workflow. |
Provenance: Frozen Batch 48 amazon fixture: module 3 observation and calculation. First-party evidence checked 2026-08-14; unresolved joins fail closed. Surface verification date: 2026-08-14. AWS Bedrock User Guide (first-party source)
Try amazon through the Batch 48 route →