Amazon

Amazon Nova 2 Lite

Amazon

Amazon Nova 2 Lite is a proprietary Amazon foundation model. A reasoning model aimed at fast, cost-effective assistants and agents. It accepts text, images, video, and documents, with configurable reasoning effort and built-in web grounding and code execution through supported services. It is generally available through Amazon Bedrock and supports a one-million-token context. Amazon has not published the model’s parameter count or detailed network architecture. Responses can contain unsupported claims, and preview/service availability varies by region and customer access.

Amazon Nova 2 Omni

Amazon

Amazon Nova 2 Omni is a proprietary Amazon foundation model. Amazon’s unified multimodal Nova 2 model accepts text, image, video, and audio and generates text and images. It combines cross-modal reasoning with a one-million-token context for document, audio, and video work. The first offering is an early-access preview for Nova Forge customers. The underlying architecture and parameter count are not public. Amazon has not published the model’s parameter count or detailed network architecture. Responses can contain unsupported claims, and preview/service availability varies by region and customer access.

Amazon Nova 2 Pro

Amazon

Amazon Nova 2 Pro is a proprietary Amazon foundation model. The larger Nova 2 reasoning tier targets difficult planning, code generation, and complex multi-step tasks. It was released in preview through Nova Forge early access, alongside generally available Nova 2 Lite. This is a controlled customer preview rather than a roadmap-only announcement. Architecture and parameter count are not disclosed. Amazon has not published the model’s parameter count or detailed network architecture. Responses can contain unsupported claims, and preview/service availability varies by region and customer access.

Amazon Nova 2 Sonic

Amazon

Amazon Nova 2 Sonic is a proprietary Amazon foundation model. A speech-to-speech dialogue model that integrates spoken understanding and response generation for natural conversation. It supports multilingual interactions, tool integration, and turn-taking behavior for real-time voice applications. It is distinct from a transcription-only service. March and May 2026 updates refresh the same serving model rather than introducing a separately named generation. Amazon has not published the model’s parameter count or detailed network architecture. Responses can contain unsupported claims, and preview/service availability varies by region and customer access.

Amazon Nova Premier

Amazon

Amazon Nova Premier is a proprietary Amazon foundation model. The highest-capability first-generation Nova tier handles complex multimodal understanding, large document collections, and specialist tasks. It accepts text, images, and video and can serve as a teacher for Bedrock model distillation into lower-cost Nova models. It has a one-million-token context and is a proprietary service model, separate from the earlier Micro, Lite, and Pro tiers. Amazon has not published the model’s parameter count or detailed network architecture. Responses can contain unsupported claims, and preview/service availability varies by region and customer access.

Amazon Nova Sonic

Amazon

Amazon Nova Sonic is a proprietary Amazon foundation model. The initial Nova spoken-dialogue model unifies speech recognition, contextual language understanding, and speech generation. It targets conversational assistants with natural prosody, interruptibility, and tool use, rather than simply converting speech to transcripts. The model is provided through Amazon Bedrock streaming interfaces. Nova 2 Sonic is a later, separately named generation. Amazon has not published the model’s parameter count or detailed network architecture. Responses can contain unsupported claims, and preview/service availability varies by region and customer access.

Amazon Nova (Pro)

Amazon

Family of models (Nova Micro, Lite, Pro) launched as part of AWS's Bedrock services. Details are limited, but these models power Amazon's own generative AI offerings, similar to how Azure uses OpenAI models.

AlexaTM 20B

Amazon

20B-parameter seq2seq model (teacher-student ensemble) for powering Amazon Alexa's multilingual understanding. Uses a bidirectional encoder-decoder architecture for translation and dialogue tasks.