TheRouter.aiTheRouter.ai
Back to Models

Amazon Nova Micro

amazonamazon/nova-micro

Amazon's fastest and most cost-effective text-only model. Ideal for high-throughput, low-latency tasks.

Amazon Bedrock model card ↗Nova responsible AI service card ↗Amazon Nova product page ↗

Amazon Nova Micro is the fastest and lowest-cost text-only model in the first-generation Nova family. AWS positions it for summarization, translation, classification, short completions, and other high-throughput text jobs where latency and unit cost matter more than multimodal understanding.

For TheRouter operators, Nova Micro is a routing floor: keep it for cheap text classification, routing decisions, entity extraction, and fallback chat with a 128K context window. It is not the right default for images, video, structured-output guarantees, or deep reasoning. If those are requirements, move up to Nova Lite / Pro or a non-Amazon model rather than over-promising what Micro exposes on Bedrock.

Best for
  • • High-volume text classification and routing where a cheap model can decide intent, policy, or destination before escalating
  • • Short summarization, translation, entity extraction, and normalization over text-only inputs
  • • Cost-controlled fallback chat for workloads that fit within 128K context and do not need vision or audio
  • • Few-shot classification and templated generation inside AWS / Bedrock-bound enterprise stacks
Reach for something else if
  • • Vision, document-image, video, or audio workflows — Nova Micro is text input and text output only
  • • Strict structured-output pipelines that require Bedrock native structured outputs; AWS marks that feature unsupported for Micro
  • • Deep reasoning, long agent loops, or frontier coding — route to Nova Pro / Premier, Claude, DeepSeek, or OpenAI reasoning models

How TheRouter serves this differently from the vendor

As the vendor operates it

AWS serves Nova Micro through Bedrock bedrock-runtime with Converse, Invoke, response streaming, guardrails, client-side tool calling, and prompt caching. The native Bedrock card marks structured outputs unsupported and lists text as the only input and output modality.

On TheRouter

TheRouter exposes the same text-only model as OpenAI-compatible /v1/chat/completions under amazon/nova-micro and currently routes it through the US geo Bedrock profile. TheRouter does not make Bedrock structured outputs or non-text modalities available for this model.

docs.aws.amazon.com/bedrock ↗
Context Length
128K
Max Output
5K
Input Priceper 1M tokens
$0.0378/ 1M tokens
Output Priceper 1M tokens
$0.1512/ 1M tokens

Modalities

text→text

Pricing Breakdown

TypeRate
Input$0.0378 / 1M tokens
Output$0.1512 / 1M tokens

Supported Parameters

temperaturemax_tokenstop_ptoolstool_choiceresponse_formatstop

Specifications

Launch date2024-12-05docs.aws.amazon.com/bedrock ↗verified
LifecycleActive; EOL no sooner than 2025-12-04docs.aws.amazon.com/bedrock ↗verified
Knowledge cutoffOctober 2024docs.aws.amazon.com/bedrock ↗verified
Bedrock model idamazon.nova-micro-v1:0docs.aws.amazon.com/bedrock ↗verified
TheRouter upstream profileus.amazon.nova-micro-v1:0docs.aws.amazon.com/bedrock ↗verified
Prompt cachingSupported; 1K minimum per checkpoint, up to 4 checkpoints, 5 minute TTLdocs.aws.amazon.com/bedrock ↗verified
Unsupported Bedrock featureNative structured outputsdocs.aws.amazon.com/bedrock ↗verified

Benchmarks

BenchmarkDistributionScoreSource
Public benchmark disclosure
The Bedrock model card documents context, modalities, lifecycle, and feature support, but does not publish per-benchmark scores for Nova Micro.
—Not publicly disclosed—
Nova family service card
AWS says Nova Micro delivers the lowest latency responses and lowest cost per inference in the Nova family, but the service card does not provide a numeric latency table.
Qualitative onlydocs.aws.amazon.com/ai/responsible-ai ↗
Latency / throughput ledger
No operator-funded live performance run is recorded for this page. Keep latency claims qualitative until a verification ledger exists.
—Not publicly disclosed—

API Usage Examples

Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.

cURL
curl https://api.therouter.ai/v1/chat/completions   -H "Content-Type: application/json"   -H "Authorization: Bearer $THE_ROUTER_API_KEY"   -d '{
    "model": "amazon/nova-micro",
    "messages": [
      {"role": "user", "content": "Summarize the key points from this input."}
    ]
  }'

API guide

Full API reference →

Chat completion

Call Nova Micro through TheRouter's OpenAI-compatible chat surface for low-cost text tasks.

cURL
curl https://api.therouter.ai/v1/chat/completions \
  -H "Authorization: Bearer $THEROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "amazon/nova-micro",
    "messages": [{"role": "user", "content": "Classify this support ticket as billing, technical, or sales."}]
  }'

More from amazon

amazon/nova-liteStep up when the task needs image or video input while staying in the low-cost Nova tier.amazon/nova-proHigher-accuracy multimodal Nova for workloads where Micro's text-only speed is not enough.amazon/nova-premierMost capable first-generation Nova and a teacher model for distillation, not a cost-floor router.amazon/nova-2-liteNewer Nova 2 option with multimodal input, 1M context, and extended-thinking budgets.

Similar models

Cross-provider sibling models
openai/gpt-5-nano
OpenAI's ultra-low-cost text tier; compare when provider lock-in matters less than OpenAI ecosystem compatibility.
google/gemini-3.5-flash-lite
Google low-cost flash tier; useful when latency and price matter but you also want a non-AWS fallback.
qwen/qwen3-8b
Small Qwen alternative for cheap multilingual text tasks and self-host-adjacent routing plans.

News & changes

2024-12-05

AWS launches Amazon Nova Micro as the fastest text-only Nova model

AWS's Bedrock model card lists Nova Micro as a text-only active model with a 128K context window, 5K max output, October 2024 knowledge cutoff, and support for Converse, Invoke, response streaming, guardrails, client-side tool calling, and prompt caching.

re-authored by TheRouterdocs.aws.amazon.com/bedrock ↗
2025-04-30

Amazon updates responsible AI service card for Nova Micro, Lite, Pro, and Premier

The service card frames Nova Micro as the lowest-latency, lowest-cost member of the Nova family and describes intended text-generation, summarization, classification, RAG, and agentic use cases while reminding customers to evaluate on their own content.

re-authored by TheRouterdocs.aws.amazon.com/ai/responsible-ai ↗

Frequently asked

Is Amazon Nova Micro text-only?

Yes. AWS lists text as the only supported input and output modality for Nova Micro. Use Nova Lite, Pro, Premier, or Nova 2 Lite if the request needs images, documents-as-images, video, or audio.

docs.aws.amazon.com/bedrock ↗
What context window does Nova Micro have?

128K tokens, with up to 5K output tokens according to the Bedrock model card.

docs.aws.amazon.com/bedrock ↗
Does Nova Micro support structured outputs?

Not as a Bedrock native feature. AWS marks structured outputs as unsupported for Nova Micro. You can still ask for JSON in the prompt through TheRouter, but you should validate it in your application.

docs.aws.amazon.com/bedrock ↗
Which Bedrock model id backs TheRouter's amazon/nova-micro?

AWS lists amazon.nova-micro-v1:0 as the in-region model id and us.amazon.nova-micro-v1:0 / eu.amazon.nova-micro-v1:0 as geo inference IDs. TheRouter currently routes the public amazon/nova-micro model through the US geo profile.

docs.aws.amazon.com/bedrock ↗
Should new workloads use Nova Micro or Nova 2 Lite?

Use Nova Micro when the job is strictly text-only and price or latency dominates. Evaluate Nova 2 Lite when you need 1M context, multimodal input, or extended thinking. The older Micro model remains useful as a cost-floor router, not as a general-purpose successor path.

docs.aws.amazon.com/ai/responsible-ai ↗
Fact ledger — every claim on this page traces here
sourceURLretrieved
Launch datedocs.aws.amazon.com/bedrock ↗2026-08-08verified
Lifecycledocs.aws.amazon.com/bedrock ↗2026-08-08verified
Knowledge cutoffdocs.aws.amazon.com/bedrock ↗2026-08-08verified
Bedrock model iddocs.aws.amazon.com/bedrock ↗2026-08-08verified
TheRouter upstream profiledocs.aws.amazon.com/bedrock ↗2026-08-08verified
Prompt cachingdocs.aws.amazon.com/bedrock ↗2026-08-08verified
Unsupported Bedrock featuredocs.aws.amazon.com/bedrock ↗2026-08-08verified
Public benchmark disclosuredocs.aws.amazon.com/bedrock ↗2026-08-08unknown
Nova family service carddocs.aws.amazon.com/ai/responsible-ai ↗2026-08-08verified
Latency / throughput ledgerdocs.aws.amazon.com/bedrock ↗2026-08-08unknown
AWS launches Amazon Nova Micro as the fastest text-only Nova modeldocs.aws.amazon.com/bedrock ↗2026-08-08verified
Amazon updates responsible AI service card for Nova Micro, Lite, Pro, and Premierdocs.aws.amazon.com/ai/responsible-ai ↗2026-08-08verified
Is Amazon Nova Micro text-only?docs.aws.amazon.com/bedrock ↗2026-08-08to verify
What context window does Nova Micro have?docs.aws.amazon.com/bedrock ↗2026-08-08to verify
Does Nova Micro support structured outputs?docs.aws.amazon.com/bedrock ↗2026-08-08to verify
Which Bedrock model id backs TheRouter's amazon/nova-micro?docs.aws.amazon.com/bedrock ↗2026-08-08to verify
Should new workloads use Nova Micro or Nova 2 Lite?docs.aws.amazon.com/ai/responsible-ai ↗2026-08-08to verify
Help & contact