Last verified
CLOUD GIANTAMAZON NOVAON BEDROCK1M CONTEXTMULTIMODAL

Amazon Nova API Pricing

Amazon Nova is the family of first-party foundation models built by Amazon and served on Amazon Bedrock. The current Nova 2 generation replaced the original Nova Micro/Lite/Pro/Premier lineup, led by Nova 2 Lite at a 1M-token context window with on-demand per-token pricing.

Priced models
1
1 GA
Founded
1994
Amazon - Seattle
Cheapest input
$0.3/M
Amazon Nova 2 Lite input
Max context
1M
Amazon Nova 2 Lite
Billing
token PAYG
Bedrock on-demand
Data retention
0 days
no training on prompts
§ 01 / LINEUP

The full roster.

Side-by-side →
§ 02 / SHELF

All side-by-side.

Methodology →
Model Input /M Output /M Cached Context Max output Vision Tools Tier
Amazon Nova 2 Lite FLAGSHIP $0.3 $2.50 1M Active

The Amazon Nova story so far.

Releases and pricing from Amazon's first-party model line on Bedrock. Sourced from AWS press, the AWS News blog, and the Bedrock pricing page - verified at publication.

DEC 02 - 2025
Nova 2 Lite released - price-performance tier with a 1M-token context on Bedrock
RELEASE
DEC - 2025
Nova 2 generation announced - Lite (GA) plus Pro and Omni in preview, replacing the original Nova lineup
RELEASE
2025
Amazon adds global cross-region inference pricing - materially cheaper than single-region on-demand for Nova
PRICING
DEC - 2024
Amazon Nova launched at re:Invent 2024 - first Amazon first-party foundation models (Micro, Lite, Pro, Premier)
RELEASE
§ 04 / ACCESS

Where to get it.

Methodology →
§ 04 / BEST FOR

Which Amazon for what.

More scenarios →
High-volume multimodal apps on a budget?
Nova 2 Lite
Nova 2 Lite
Complex reasoning and agentic workflows?
Nova 2 Pro (preview)
Bedrock pricing
Any-to-any audio/video generation?
Nova 2 Omni (preview)
Amazon Nova
Cheapest input on Bedrock first-party?
Nova 2 Lite at $0.30/M
Nova 2 Lite
§ 06 / BACKGROUND

The company behind it.

aws.amazon.com/ai/generative-ai/nova →

Amazon was founded by Jeff Bezos in 1994 and is headquartered in Seattle. Its cloud division, Amazon Web Services (AWS), is the largest cloud provider and the home of Amazon's generative-AI stack.

Amazon introduced its own Amazon Nova foundation models at re:Invent in December 2024, initially as Micro, Lite, Pro and Premier. In December 2025 Amazon shipped the Nova 2 generation, led by Nova 2 Lite (generally available) with Nova 2 Pro and Nova 2 Omni in preview.

Nova models are served exclusively through Amazon Bedrock, where they are billed per token on demand. Amazon prices Nova with a split between global cross-region inference and single-region (Geo) inference, and offers Batch, provisioned-throughput and reserved options. Amazon states it does not train on customer prompts submitted through Bedrock.

Amazon also invests heavily in Anthropic and hosts third-party frontier models on Bedrock, so Nova sits alongside Claude, Llama, Mistral and others in Amazon's catalogue rather than as its only option.

§ 07 / COMPETITORS

Other frontier labs.

All providers →
CLOUD RIVAL

Google (Gemini)

Gemini on Vertex AI is the closest cloud-native rival with strong multimodal and long context, but Google's own models are pricier at the Pro tier than Nova 2 Lite.

Vertex AI
PARTNER + RIVAL

Anthropic (Claude)

Claude is Amazon's flagship third-party model on Bedrock and an Amazon investment, far stronger on frontier coding but many times more expensive than Nova 2 Lite.

on Bedrock
FRONTIER

OpenAI

OpenAI leads brand demand and frontier quality, but is not first-party on Bedrock; Nova competes on price and native AWS integration rather than raw capability.

GPT family
COST DISRUPTOR

DeepSeek

DeepSeek undercuts nearly everyone on raw token price, but Nova offers managed AWS hosting, data residency and multimodal input that a low-cost API does not.

$0.14/M in
OPEN-WEIGHT GIANT

Alibaba (Qwen)

The Qwen (Tongyi Qianwen) model family from Alibaba Cloud's Tongyi Lab. First released in 2023, Qwen is the most-downloaded open-weight model family in the world — most tiers ship under Apache 2.0 on Hugging Face and ModelScope, while the proprietary Max tier is served through Alibaba Cloud's Model Studio.

28 models · from $0.05/M
CHINA CONSUMER LEADER

ByteDance (Doubao)

The Doubao model family from ByteDance — parent of TikTok and Douyin. Served through the Volcano Ark (火山方舟) platform on ByteDance's Volcano Engine cloud, Doubao powers China's most-used consumer AI app and undercuts most frontier labs on price.

14 models · from $0.02/M

Amazon Nova pricing questions.

Practical questions about Amazon Nova API pricing on Bedrock, with the published per-token rates kept separate from workload assumptions.

Q · 01 How much does the Amazon Nova API cost? +
On Amazon Bedrock, Nova 2 Lite is $0.30/M input and $2.50/M output using global cross-region inference. Nova 2 Pro (preview) is $1.25/$10 and Nova 2 Omni (preview) is $0.30/$2.50 for text. Verified on the Bedrock pricing page 2026-07-11.
Q · 02 What happened to Nova Micro, Lite, Pro and Premier? +
Those were the original 2024 Nova generation. Amazon's current Bedrock pricing page lists the Nova 2 generation instead, so older Micro/Lite/Pro/Premier per-token numbers you may find in blogs are stale. This page tracks Nova 2.
Q · 03 Why are there two prices per model? +
Amazon lists global cross-region inference (routes your request to the cheapest available region) and single-region Geo pricing. For Nova 2 Lite, global is $0.30/$2.50 while in-region is roughly $0.525/$4.375 - about 1.75x higher. This page uses the global cross-region rate as the baseline.
Q · 04 Does Amazon train on my Nova prompts? +
No. Amazon states that content submitted to models on Bedrock is not used to train the underlying models and is not shared with model providers, with encryption in transit and at rest and standard AWS data-residency controls.
Q · 05 Is there a batch discount? +
Yes. Amazon Bedrock offers a Batch mode at roughly 50% of on-demand token price for latency-tolerant jobs, plus provisioned-throughput and reserved options for high, steady volume.