Amazon Nova 2 Lite

vNova 2 Lite (GA)

Amazon (AWS)

Modelawsbedrockenterprisehipaa-eligible
87
Strong
About This Model

Amazon's cost-efficient Nova 2 workhorse model, GA on Amazon Bedrock since re:Invent 2025. Offers three thinking-intensity levels, a built-in code interpreter, and web grounding with a 1M token context, backed by AWS's strong enterprise compliance posture.

Last Evaluated: June 10, 2026
Official Website

Trust Vector Analysis

Dimension Breakdown

🚀Performance & Reliability
+

GA member of the Nova 2 family (Nova 2 Pro and Omni remain in preview). Public benchmark data is still sparse, so performance scores carry medium/low confidence.

task accuracy code

Review of provider claims; limited independent benchmark coverage to date

Evidence
AWS What's New (Nova 2 announcement)Built-in code interpreter enables executable code workflows; AWS cites competitive coding quality for its price tier
mediumVerified: 2026-06-10
task accuracy reasoning

Review of provider evaluations of adjustable-thinking reasoning performance

Evidence
AWS What's New (Nova 2 announcement)Three thinking-intensity levels allow scaling reasoning depth per request
mediumVerified: 2026-06-10
task accuracy general

Review of provider documentation; independent leaderboard coverage remains sparse

Evidence
Amazon Nova DocumentationPositioned as fast, capable general-purpose model; few public leaderboard placements
lowVerified: 2026-06-10
output consistency

Review of feature design and repeated-prompt behavior reports

Evidence
Amazon Nova DocumentationThinking-intensity controls produce predictable quality/cost tradeoffs
mediumVerified: 2026-06-10
latency p50

Median latency from third-party benchmarking at default settings

Evidence
Community benchmarkingLite tier optimized for low latency; higher thinking intensity increases response time
mediumVerified: 2026-06-10
latency p95

95th percentile estimates across configurations

Evidence
Community benchmarkingTail latency dominated by thinking-intensity setting and tool use (code interpreter, web grounding)
lowVerified: 2026-06-10
context window

Official specification from provider documentation

Evidence
Amazon Nova Documentation1M token context window with approximately 65K max output tokens
highVerified: 2026-06-10
uptime

Historical service availability from AWS status reporting

Evidence
AWS Health DashboardBedrock service availability backed by AWS regional infrastructure and SLAs
highVerified: 2026-06-10
🛡️Security
+

Inherits AWS's mature platform security: IAM, PrivateLink, KMS, CloudTrail, and Bedrock Guardrails give it one of the strongest deployment security stories available.

prompt injection resistance

Testing against OWASP LLM01 patterns with and without Bedrock Guardrails

Evidence
Amazon Bedrock GuardrailsBedrock Guardrails add configurable prompt-attack filtering on top of model defenses
mediumVerified: 2026-06-10
jailbreak resistance

Review of adversarial testing and platform guardrail capabilities

Evidence
AWS Responsible AISafety training plus platform-level guardrails reduce jailbreak success
mediumVerified: 2026-06-10
data leakage prevention

Analysis of Bedrock data protection documentation and commitments

Evidence
Amazon Bedrock Data ProtectionCustomer data not used to train models; content not shared with model providers
highVerified: 2026-06-10
output safety

Safety testing across harmful content categories

Evidence
AWS Responsible AIBuilt-in content moderation plus configurable Bedrock Guardrails policies
mediumVerified: 2026-06-10
api security

Review of AWS-native security controls available to Bedrock workloads

Evidence
Amazon Bedrock SecurityIAM authentication, SigV4 signing, VPC endpoints (PrivateLink), KMS encryption, CloudTrail auditing
highVerified: 2026-06-10
🔒Privacy & Compliance
+

Among the strongest compliance postures in the market via Bedrock: SOC, ISO, HIPAA-eligible, GDPR-supporting, with data never used for training.

data residency

Review of Bedrock regional processing documentation

Evidence
Amazon Bedrock DocumentationProcessing stays in selected region (cross-region inference configurable)
highVerified: 2026-06-10
training data optout

Analysis of Bedrock data usage commitments

Evidence
Amazon Bedrock Data ProtectionCustomer prompts and outputs are never used to train Nova models — no opt-out required
highVerified: 2026-06-10
data retention

Review of Bedrock retention and logging documentation

Evidence
Amazon Bedrock Data ProtectionBedrock does not store prompts/outputs after processing unless customer enables invocation logging
highVerified: 2026-06-10
pii handling

Review of platform PII detection and redaction capabilities

Evidence
Amazon Bedrock GuardrailsGuardrails support sensitive-information filters including PII redaction
mediumVerified: 2026-06-10
compliance certifications

Verification against AWS services-in-scope compliance listings

Evidence
AWS Compliance ProgramsBedrock in scope for SOC 1/2/3, ISO 27001/27017/27018, HIPAA-eligible, GDPR-supporting, FedRAMP
highVerified: 2026-06-10
zero data retention

Review of data handling practices

Evidence
Amazon Bedrock Data ProtectionNo persistent storage of inference content by default; abuse-detection processing is ephemeral
highVerified: 2026-06-10
👁️Trust & Transparency
+

Strong platform-level guardrails and grounding features; independent transparency and factuality data still limited for the Nova 2 generation.

explainability

Evaluation of reasoning transparency and grounded citation behavior

Evidence
AWS What's New (Nova 2 announcement)Thinking-intensity levels expose reasoning effort; web grounding cites sources
mediumVerified: 2026-06-10
hallucination rate

Review of grounding features; independent factual QA coverage sparse

Evidence
Amazon Nova DocumentationBuilt-in web grounding designed to reduce hallucinations; limited independent factuality data
lowVerified: 2026-06-10
bias fairness

Review of AWS responsible AI documentation and service cards

Evidence
AWS Responsible AIAWS responsible AI program with published service cards and bias testing
mediumVerified: 2026-06-10
uncertainty quantification

Qualitative assessment of confidence expression

Evidence
Model BehaviorAdequate uncertainty expression; little published evaluation
lowVerified: 2026-06-10
model card quality

Review of documentation completeness

Evidence
Amazon Nova DocumentationAWS publishes service cards and user guides covering capabilities and responsible-use guidance
mediumVerified: 2026-06-10
training data transparency

Review of public disclosures about training data

Evidence
AWS Public StatementsTraining data sources not disclosed in detail
mediumVerified: 2026-06-10
guardrails

Analysis of built-in and platform-level safety mechanisms

Evidence
Amazon Bedrock GuardrailsConfigurable platform guardrails: content filters, denied topics, PII redaction, contextual grounding checks
highVerified: 2026-06-10
⚙️Operational Excellence
+

Exceptional operational maturity by virtue of the AWS/Bedrock platform: IAM, monitoring, lifecycle policy, and enterprise support are best-in-class.

api design quality

Review of API design and feature completeness

Evidence
Amazon Bedrock APIConverse API with streaming, tool use, thinking-intensity controls, code interpreter, web grounding
highVerified: 2026-06-10
sdk quality

Review of SDK quality, documentation, and maintenance

Evidence
AWS SDKsMature official SDKs across Python, JavaScript, Java, Go, .NET and more
highVerified: 2026-06-10
versioning policy

Review of Bedrock model lifecycle policy

Evidence
Amazon Bedrock Model LifecycleDocumented model lifecycle with legacy/EOL stages and advance notice
mediumVerified: 2026-06-10
monitoring observability

Review of monitoring and observability integrations

Evidence
Amazon CloudWatch / CloudTrailNative CloudWatch metrics, invocation logging, and CloudTrail audit trails
highVerified: 2026-06-10
support quality

Assessment of support tiers and documentation

Evidence
AWS SupportTiered enterprise support with TAMs, SLAs, and extensive documentation
highVerified: 2026-06-10
ecosystem maturity

Analysis of platform integrations and tooling

Evidence
Amazon Bedrock EcosystemIntegrates with Bedrock Agents, Knowledge Bases, Guardrails, and the broader AWS ecosystem
highVerified: 2026-06-10
license terms

Review of licensing terms and indemnification

Evidence
AWS Service TermsStandard AWS service terms; Amazon offers IP indemnification for Nova outputs
highVerified: 2026-06-10
Strengths
  • +Very low pricing: $0.30/$2.50 per 1M tokens with 1M context
  • +Three thinking-intensity levels for per-request quality/cost control
  • +Built-in code interpreter and web grounding (no external tooling needed)
  • +Best-in-class enterprise compliance via Bedrock (SOC, ISO, HIPAA-eligible, GDPR)
  • +Customer data never used for model training
  • +Deep AWS integration: IAM, PrivateLink, KMS, CloudWatch, Guardrails
Limitations
  • !Public benchmark data sparse; performance claims rest largely on AWS evaluations
  • !Raw capability trails frontier flagships on hard reasoning and coding
  • !Nova 2 Pro and Omni siblings still in preview, limiting family upgrade paths
  • !AWS-only availability creates platform lock-in
  • !Training data transparency is limited
Metadata
pricing
input: $0.30 per 1M tokens
output: $2.50 per 1M tokens
notes: Amazon Bedrock pricing; batch and provisioned throughput options available. Announced at re:Invent 2025-12-02.
last verified: 2026-06-10
context window: 1000000
max output: 65000
languages
0: English
1: Spanish
2: French
3: German
4: Italian
5: Portuguese
6: Japanese
7: Korean
8: Chinese
9: Arabic
10: Hindi
modalities
0: text
1: image (input)
2: document
api endpoint: https://bedrock-runtime.us-east-1.amazonaws.com
open source: false
architecture: Transformer-based with adjustable thinking intensity, built-in code interpreter, and web grounding
parameters: Not disclosed
release date: 2025-12-02 (GA at re:Invent; Nova 2 Pro and Omni in preview)

Use Case Ratings

code generation

Built-in code interpreter is convenient for executable workflows, but raw coding quality trails frontier flagships.

customer support

Low cost, low latency, 1M context, and strong compliance make it excellent for enterprise support at scale.

content creation

Capable everyday content generation with web grounding for current facts; less distinctive prose than flagships.

data analysis

Code interpreter plus 1M context suit data workflows well at a very low price point.

research assistant

Web grounding with citations and 1M context are strong for research at this price tier.

legal compliance

Bedrock's compliance certifications and data controls are the draw; pair with human review for analysis depth.

healthcare

HIPAA-eligible on Bedrock with strong data protections; validate clinical accuracy given sparse benchmarks.

financial analysis

Good quantitative workflows via code interpreter; compliance posture fits financial services.

education

Adjustable thinking intensity and low cost suit high-volume tutoring deployments.

creative writing

Serviceable but not a creative standout versus frontier flagships.