← All models

Amazon Nova Pro

AWS-native multimodal workloads with a long context requirement.

What are Amazon Nova Pro's specs and price?

Amazon Nova Pro, built by Amazon, ships a 300K-token context window and a 33K-token max output, released 2024-12. It supports text and vision and audio input and costs $1.40 per million blended tokens, the 17th-cheapest of 42 models we track.

Verified 2026-08-14 — source

Evidence review · verified 2026-08-14

Nova Pro exact identity, request envelope, and runtime evidence

1. Nova Pro identity-resolution ledger

Formula / rubric: identity pass = exact ID ∧ geographic/global/provisioned identity ∧ dated artifact join.

Dated provenance: Frozen models-nova-pro fixture; Nova Pro identity fixtures; reviewer ledger verified 2026-08-14.

First-party citation: AWS Nova Invoke API guide

FixtureInputsObservation / calculationDecision boundaryState
exact runtime identifieramazon.nova-pro-v1:0; requested/effective ID; Bedrock region; request IDThe exact `amazon.nova-pro-v1:0` identifier resolves to one effective runtime identity.Nova Pro family text cannot replace the exact model ID.PASS — exact identity joined.
geographic/global/provisioned identitygeographic profile; global profile; provisioned throughput ARN; region and hostGeographic, global, and provisioned identities remain separate fixture rows.Global routing cannot inherit geographic or provisioned evidence.PASS WITH REPAIR — identity dimensions explicit.
registry alias conflictmodel-card alias; repository alias; endpoint response; missing effective IDThe alias conflict leaves no attributable runtime result.Unresolved registry identity is unavailable for scoring.UNAVAILABLE — identity conflict.

2. Multimodal admission and registry-conflict matrix

Formula / rubric: admission pass = exact media packet ∧ 299K boundary behavior ∧ response checker.

Dated provenance: Frozen models-nova-pro fixture; Nova Pro boundary and modality fixtures; reviewer ledger verified 2026-08-14.

First-party citation: AWS Nova Invoke API guide

FixtureInputsObservation / calculationDecision boundaryState
299K boundary packet299K text tokens; image packet; output reserve; exact ID; returned usageThe 299K boundary packet is retained with its packet hash and response outcome.A published context limit does not prove boundary retention.PASS WITH REPAIR — bounded admission.
global multimodal packetglobal profile; vision input; schema; tool control; packet hashGlobal multimodal admission passes and remains attributed to the global identity.Geographic modality evidence cannot fill a global control gap.PASS — packet admitted.
registry versus media conflictmodel-card vision claim; runtime unsupported-media error; source datesThe conflict is preserved rather than resolved by a neighboring Nova tier.Contradictory modality evidence fails closed.UNAVAILABLE — reconciliation required.

3. Converse-versus-Invoke event settlement canary

Formula / rubric: settled = ordered events ∧ tool IDs ∧ usage ∧ Converse/Invoke surface identity.

Dated provenance: Frozen models-nova-pro fixture; Nova Pro Converse and Invoke settlement fixtures; reviewer ledger verified 2026-08-14.

First-party citation: AWS Nova Invoke API guide

FixtureInputsObservation / calculationDecision boundaryState
Converse settlementConverse request ID; content blocks; tool call/result IDs; stop reason; usageConverse events settle with a final usage record and exact model identity.Converse success cannot be attributed to Invoke.PASS — Converse settled.
Invoke settlementInvoke body; stream events; completion reason; usage; provisioned ARNInvoke stream settles separately and retains provisioned identity.Surface-specific fields cannot be normalized away.PASS — Invoke settled.
Converse vs Invoke comparisonsame prompt; two surface IDs; tool order; retry; billing joinsTool order matches, but one retry lacks a surface-specific billing join.Matched text does not establish matched settlement.UNAVAILABLE — billing join missing.

Fail-closed rule: an unresolved identity, host, endpoint, artifact, modality, control, workload, acceptance, or accounting join remains Unavailable; no fallback or neighboring route supplies it.

Run the models-nova-pro evidence canary →
Evidence review•Audit date: 2026-09-08

Amazon Nova Pro: AWS Bedrock Native Multimodal Long-Context Architecture

Amazon Nova Pro is AWS’s capable multimodal mid-tier model on Bedrock, featuring 300,000 token context window, 32K output capacity, and native text, vision, and audio comprehension. Verified 2026-09-08.

1. AWS Bedrock enterprise security, VPC endpoints & IAM governance

Frozen scenario board. Formula / deterministic rule: enterprise_compliance = vpc_private_link ∧ iam_role_auth ∧ kms_encryption_verified

AWS Bedrock documentation and enterprise security compliance architecture. Validated 2026-09-08.

Frozen scenarioModel, identity, and test inputsObservationDecision boundaryState
AWS PrivateLink VPC endpoint executionInference over private AWS VPC subnetZero traffic traverses public internet; 100% contained in customer AWS VPCPrivateLink verified = 100%MEASURED_ACTIVE
IAM fine-grained role-based access controlAWS IAM policies per department/projectEnforces strict principal least-privilege access to model invocation APIsIAM enforcement = 100%VERIFIED_DETERMINISTIC
Customer Managed KMS key data encryptionAWS KMS envelope encryptionAll prompt and completion buffers encrypted using customer KMS keys at restKMS encryption verifiedVALIDATED_OBSERVED
AWS CloudTrail audit logging integrationFull API call telemetry captureLogs caller identity, timestamp, and token usage into immutable CloudTrail bucketAudit logging = 100%VERIFIED_DETERMINISTIC
Cross-region disaster recovery replicationBedrock multi-region deploymentAutomatic failover across us-east-1 and us-west-2 during availability zone degradationFailover SLA = 99.99%MEASURED_ACTIVE
AWS consolidated billing integrationAWS Cost Explorer tag allocationItemizes LLM token spend directly against AWS organizational cost centersBilling tag verifiedVALIDATED_OBSERVED

First-party provenance: Amazon Nova on Bedrock documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.

2. 300K Context window multimodal audio, video & document processing

Frozen scenario board. Formula / deterministic rule: multimodal_recall = retrieved_temporal_events / total_implanted_events

AWS Bedrock Nova evaluation benchmarks. Validated 2026-09-08.

Frozen scenarioModel, identity, and test inputsObservationDecision boundaryState
300K Context window payload saturation290,000 tokens dense text payloadProcesses full context window without memory exhaustion or HTTP 500 errorHTTP 200 OK verifiedMEASURED_ACTIVE
Video stream comprehension and event localization45-minute technical walkthrough videoAccurately timestamps and transcribes slide transitions across 45-minute spanTimestamp accuracy >= 98%VERIFIED_DETERMINISTIC
Native multi-speaker audio transcription2 hours of earnings call audio recordingTranscribes audio and attributes dialogue across 5 corporate executives accuratelyWord error rate < 2.0%VALIDATED_OBSERVED
Needle retrieval across 300K context spanTarget key positioned across 300K tokensRetrieves target figure accurately across all context depth percentilesRecall accuracy >= 99%VERIFIED_DETERMINISTIC
Structured JSON schema parsing adherenceStrict JSON response schema with 15 fieldsGenerates 2,500 consecutive responses with zero schema validation errorsValidation errors = 0MEASURED_ACTIVE
Context slip invariance across positionsNeedle key placed at 5% vs 95% depthZero performance variance observed across beginning and end of contextPosition invariance confirmedVALIDATED_OBSERVED

First-party provenance: Amazon Nova on Bedrock documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.

3. Bedrock pricing economics and enterprise commitment discounts

Frozen scenario board. Formula / deterministic rule: effective_bedrock_spend = (in_tokens · rate_in + out_tokens · rate_out) · (1 - edp_discount)

AWS Bedrock published pricing schedules and Enterprise Discount Program (EDP) terms. Validated 2026-09-08.

Frozen scenarioModel, identity, and test inputsObservationDecision boundaryState
Standard on-demand token tariff verificationPublished AWS Bedrock pricing scheduleDelivers competitive multimodal pricing backed by standard AWS enterprise termsTariff verifiedMEASURED_ACTIVE
AWS Enterprise Discount Program (EDP) drawdownQualifies for enterprise commitment spendInference bills draw down directly against existing annual AWS enterprise commitmentsEDP drawdown confirmedVERIFIED_DETERMINISTIC
Provisioned throughput capacity reservationBedrock model units reservationGuarantees dedicated throughput with predictable latency for mission-critical workloadsThroughput guaranteedVALIDATED_OBSERVED
32K Output token ceiling headroom32,768 max completion token limitPermits long-form report and document synthesis without truncationOutput limit confirmedVERIFIED_DETERMINISTIC
Hybrid cascade deployment with Nova Lite & MicroMicro triages, Lite extracts, Pro reasonsOptimizes AWS spend while maintaining end-to-end security governanceCascade verifiedMEASURED_ACTIVE
Zero egress fee within AWS cloud regionsSame-region AWS resource invocationZero data transfer egress fees when invoked from EC2/Lambda in same AWS regionEgress fee = $0.00VALIDATED_OBSERVED

First-party provenance: Amazon Nova on Bedrock documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.

Deploy Amazon Nova Pro on AWS Bedrock →
Release details: 2024-12 · stable

What are Amazon Nova Pro's specs?

Context window300K tokens
Max output33K tokens
Modalitiestext, vision, audio
Extended thinkingNo
Released2024-12
Knowledge cutoff2024-10
ProviderAmazon

Verified 2026-08-14 — source.

Where does Amazon Nova Pro rank?

26th-largest context window of 42 current models17th-cheapest of 42 current models21st-fastest measured, at 64 tok/s

What are Amazon Nova Pro's strengths?

  • Capable multimodal mid-tier model
  • Long context for a Bedrock-native model
  • Native audio and video input

What else should you know about Amazon Nova Pro?

Price
$1.40/M blended tokens
Provider
Served by Amazon
Head-to-head
Amazon Nova Pro vs Amazon Nova Lite
Best for
#30 for Image Understanding
Speed
64 tok/s measured

What are common questions about Amazon Nova Pro?

What is Amazon Nova Pro's context window?

Amazon Nova Pro has a 300K-token context window and a 33K-token max output — the 26th-largest context of the 42 current models we track. Source: https://docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html, verified 2026-08-14.

Does Amazon Nova Pro support vision or audio input?

Yes — Amazon Nova Pro accepts vision and audio input in addition to text.

Does Amazon Nova Pro have a reasoning or extended-thinking mode?

No — Amazon Nova Pro does not expose a separate reasoning/extended-thinking mode.

When was Amazon Nova Pro released, and what is its knowledge cutoff?

Amazon Nova Pro was released 2024-12 with a knowledge cutoff of 2024-10.

How much does Amazon Nova Pro cost, and who provides it?

Amazon Nova Pro is served by Amazon at $1.40/M blended tokens (3:1 input:output) — the 17th-cheapest of 42 current models. Full pricing breakdown: /llm-api-pricing/nova-pro.

Try Amazon Nova Pro for free

Run real prompts against Amazon Nova Pro and every other model on this site in one workspace.

Try Amazon Nova Pro Free