#17Amazon

Amazon Nova 2 Pro (Preview)

Amazon Nova 2 Pro is the most capable model in Amazon's second-generation Nova family, aimed at complex agent tasks such as multi-document analysis, video reasoning and software migrations. It launched as a preview with early access for Nova Forge customers. Amazon says it equals or beats Claude Sonnet 4.5 on 10 of 16 benchmarks.

• Text, image, video and speech input
• Extended thinking at low, medium or high intensity
• 1M-token context window (AWS)
• Built-in code interpreter, web grounding and remote MCP tools

AccessRestricted preview: early access for all Amazon Nova Forge customers, via Amazon Bedrock

Amazon Nova 2 Lite

Amazon Nova 2 Lite is a fast, low-cost reasoning model for everyday work such as customer-service chat, document processing and business automation. It reads text, images and video, has adjustable extended thinking and built-in tools. Amazon says it equals or beats Claude Haiku 4.5 on 13 of 15 benchmarks.

• Text, image and video input
• 1M-token context window
• Adjustable thinking: low, medium, high
• Built-in code interpreter, web grounding and remote MCP tools

AccessGenerally available in Amazon Bedrock (global cross-region inference); customisable through Nova Forge

Amazon Nova 2 Omni (Preview)

Amazon Nova 2 Omni is a single model that takes text, images, video and speech as input and produces both text and images; Amazon called this combined text-and-image generation an industry first. It launched in early access for Nova Forge customers. Reports in July 2026 say Amazon is winding it down.

• Inputs: text, images, video and speech
• Outputs: text and images from one model
• Handles up to 750,000 words, hours of audio (Amazon)

AccessRestricted preview: early access for Amazon Nova Forge customers

Amazon Nova 2 Sonic

Amazon Nova 2 Sonic is Amazon's second-generation speech-to-speech model for real-time voice conversations such as customer-service calls. It supports seven languages, with voices that can switch language mid-conversation, natural turn-taking and handling of interruptions. Amazon updated it in place in March and May 2026 to cut speech errors and latency.

• Real-time speech-to-speech conversation
• English, French, Italian, German, Spanish, Portuguese, Hindi
• 1M-token context window (Amazon)
• May 2026 refresh: 88% fewer speech hallucinations (internal set)

AccessGenerally available in Amazon Bedrock (model ID amazon.nova-2-sonic-v1:0; US East, US West (Oregon), Asia Pacific (Tokyo)); Amazon Connect and telephony integrations

Amazon Nova Act

Amazon Nova Act is an AWS service for building and running AI agents that carry out tasks in web browsers, such as updating CRM records or testing websites. It became generally available at re:Invent 2025. Amazon reports 90% reliability on early customer workflows, and press reports say it is among the Nova products Amazon is keeping.

• Browser and UI automation agents
• Build, deploy and manage fleets of agents on AWS
• 90% reliability on early customer workflows (Amazon)

AccessGenerally available AWS service for browser agents

Price$4.75 per agent hour

Amazon Nova Multimodal Embeddings

Amazon Nova Multimodal Embeddings turns text, documents, images, video and audio into vectors in one shared space, for search and retrieval-augmented generation. AWS describes it as the first single embedding model to cover all of those content types, removing the need for a separate embedding model per format.

• Text, document, image, video and audio inputs
• Inputs up to 8K tokens; 30-second audio/video segments
• Several output embedding sizes
• Synchronous and asynchronous APIs

AccessGenerally available in Amazon Bedrock, US East (N. Virginia); synchronous and asynchronous APIs