Amazon Nova
AWS's frontier model family — text, image, video, and voice AI built for enterprise scale on Amazon Bedrock.
Overview
Amazon Nova is AWS's foundation-model family on Bedrock — Micro/Lite/Pro (text/multimodal), Canvas/Reel (image/video), Sonic (speech). Nova 2 (Lite, Pro, Omni) launched Dec 2025 with native reasoning; original flagship Nova Premier is now Legacy, retiring Sept 2026.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 11 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- Text-only
- Lowest latency
- No multimodal input
- Multimodal input
- Low-cost
- 1M context
- Extended thinking
- 1M context
- Web grounding tool
- Multimodal
- Higher accuracy
- Model distillation source
- 1M context
- Being replaced by Nova 2 Pro
- US regions only
Use cases
What you can produce with Amazon Nova
- A production-ready text or multimodal AI chatbot integrated into an AWS application via the Bedrock Converse API
- Automated summaries and structured extractions from PDF documents, financial reports, or long-form video content
- Real-time AI voice agent capable of natural spoken conversation in multiple languages, deployable over a telephony or web interface
- AI-generated product images (Nova Canvas) or short promotional videos (Nova Reel) at configurable quality tiers
- A fine-tuned 'Novella' model variant trained on proprietary enterprise data via Nova Forge, hosted on Bedrock
- Agentic workflows that autonomously call external APIs, browse data sources, and execute multi-step business tasks using Nova 2 Pro
- Batch inference pipeline processing thousands of documents or prompts at 50% cost reduction using Bedrock's batch tier
ASEAN Perspective
Amazon Nova in Southeast Asia
Amazon Nova's understanding models (Micro, Lite, Pro) have been available via cross-region inference profiles in Singapore, Tokyo, Seoul, Mumbai, and Sydney since February 27, 2025, allowing ASEAN teams to keep inference within the Asia Pacific geography and satisfy many data-residency requirements. The Singapore (ap-southeast-1) and Jakarta (ap-southeast-3) AWS regions provide local data residency anchors for PDPA, OJK, and MAS regulatory contexts. However, Nova Premier and Nova Canvas/Reel remain US-region-only as of mid-2026, limiting ASEAN enterprises that need the highest-capability tier or creative generation while staying within jurisdictional boundaries. AWS's strong public-sector presence in ASEAN (Singapore GovTech partnerships, Indonesia digital economy programmes) gives Nova a distribution advantage that pure-play AI vendors lack in the region.
Nova's price-performance case is real: Micro and Lite are among the cheapest capable models on any major cloud, and having text, image, video, and speech models under one Bedrock bill is still unmatched by a single provider. Nova 2 Pro closes the reasoning gap Amazon needed to close, beating Claude Sonnet 4.5 on a chunk of Amazon's own benchmark set, and Nova 2 Omni's unified text-and-image generation with a 1M-token window is a genuine differentiator.
Two things changed that matter for evaluation. Nova Premier — the tier this entry used to lean on for "complex reasoning" — is now marked Legacy and retires in September 2026, superseded by Nova 2 Pro. And APAC access improved: Micro, Lite, and Pro now run directly in Tokyo, Seoul, Mumbai, Singapore, and Sydney, though Nova 2 and Premier both remain US-only for now. Everything still requires Bedrock and IAM setup — there's no simple API-key onboarding.
What people say
Amazon quietly retired its own headline model. Nova Premier — the 1M-token reasoning tier Amazon leaned on to answer Gemini and Claude at launch — is now marked Legacy on Bedrock's model catalog with an end-of-life date of September 14, 2026, replaced by Nova 2 Pro. That's the biggest change since this family launched at re:Invent 2024, and it's easy to miss if you're going off older marketing material.
The price-performance pitch, on the other hand, holds up. Nova Micro runs roughly $0.035 per million input tokens and Nova Lite about $0.06 — genuinely cheap next to Claude- or GPT-class models — while the whole family (text, image via Canvas, video via Reel, speech via Sonic) bills through one Bedrock relationship. Developer feedback on G2 and in AWS forums consistently cites this ecosystem integration — IAM, Guardrails, VPC endpoints — as the draw for teams already committed to AWS.
Regional access improved meaningfully in 2026: Micro, Lite, and Pro are now available directly in Tokyo, Seoul, Mumbai, Singapore, and Sydney, addressing a real complaint from ASEAN and APAC teams about data residency. Nova 2 and the old Premier tier remain US-region-only for now, so anyone needing the top reasoning tier in-region is still waiting.
The tradeoff hasn't changed: everything runs through Amazon Bedrock, which means IAM roles, the console or boto3, and region selection before your first API call — a heavier lift than providers offering a plain API key. At launch, independent benchmarks found Nova Premier trailing Gemini 2.5 Pro on coding and math; Nova 2 Pro's October 2025 model card claims that gap has closed, though third-party verification of the newer numbers is still catching up.
Summary of public user & expert reviews, compiled by RECATOOLS.
Notable facts
- Nova Micro, Lite, and Pro were claimed to be at least 75% cheaper than the top-performing models in their respective intelligence classes at launch.
- Nova Reel charges by the second of video generated — at $0.08/second, a one-minute clip costs roughly $4.80.
- Nova Forge, the enterprise model-customisation service, costs $100,000 per year and lets companies blend proprietary data with Nova checkpoints to create their own named model variants called 'Novellas'.
- Nova 2 Sonic supports a 1,000,000-token context window for voice conversations — enough to process roughly 750,000 spoken words in a single session.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Amazon Nova's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Amazon Nova unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Amazon Nova directly →
Spotted something out of date? Suggest an update →
Alternatives to Amazon Nova
More in LLMs & Chat
