The article also discusses the use of a new LLM plugin called llm-bedrock for accessing the models in the terminal via boto3 and the Amazon Bedrock API. The author provides a comparison of the pricing and capabilities of the Nova models with other models from different providers. The Nova models are praised for their multi-modal capabilities and competitive pricing. However, the process of gaining API access to anything on AWS is criticized as being overly complicated.
Key takeaways:
- Amazon has released three new Large Language Models called Amazon Nova in three sizes: Micro, Lite, and Pro. These models are competitive with the Google Gemini family and are extremely inexpensive.
- The Nova models can handle large context and the two larger models can handle images, video, and PDFs. However, they can't handle audio and ignore audio in video that they analyze.
- The Nova models can be accessed in the terminal via a new LLM plugin called llm-bedrock, which uses boto3 and the Amazon Bedrock API. The plugin supports LLM attachments, allowing images, video, and PDF files to be attached to a prompt.
- Amazon plans to introduce an Amazon Nova speech-to-speech model and a model with native multimodal-to-multimodal capabilities in 2025. The most capable multimodal model, Amazon Nova Premier, is still in training and is expected to be available in early 2025.