Amazon Web Services said Amazon Nova Premier, the most capable model in the cloud provider's family of in-house large language models, is generally available.
Nova Premier is the headliner of the Nova set of models AWS announced at AWS re:Invent. Nova Premier is designed as a teacher for model distillation and can improve the other Nova models in the AWS family.
Cloud providers are increasingly leveraging their own models as LLMs go commodity. Think of these in-house models as the equivalent of private label brands at retailers. AWS has Nova. Google Cloud has Gemini. Microsoft Azure just launched a set of new Phi models on Hugging Face. The strategies all rhyme as hyperscalers launch alpha models and then a set of more lightweight ones designed for specific use cases.
- AWS, Amazon court developers to take Nova models for a spin, Alexa+ goes early access
- Amazon's Alexa gets a brain transplant, showcases Amazon Bedrock, Nova
- AWS launches Amazon Nova foundation models in commoditization play
- Google Gemini vs. OpenAI, DeepSeek vs. Qwen: What we're learning from model wars
AWS said Nova Premier will be available in Amazon Bedrock. Amazon Nova Premier will be able to reason through complex tasks that require understanding of context , planning and execution across tools and data source. Nova Premier has a context length of 1 million tokens.
According to AWS, Nova Premier and Amazon Bedrock Model Distillation can be used to tailor various versions of Nova models for use case.
Based on AWS' technical report on Nova Premier, the model is an improvement on Nova Pro, but trails Claude-3.5 Sonnet-v2, Claude-3.7 Sonnet and GPT-4.5 in most categories. However, Nova Premier may be good enough for multiple use cases based on cost.
