Blog
Remember when Amazon Nova was announced at re:Invent 2024?
Well, there's been some significant movement with these Models and lots of updates as different functionality comes online. I want to share both the recent updates and how we at Mission have been putting the Nova Foundation Models to the test.
Quick Recap: What’s Amazon Nova?
ICYMI: “Amazon Nova is a new generation of state-of-the-art (SOTA) foundation models (FMs) that deliver frontier intelligence and industry-leading price-performance, available exclusively on Amazon Bedrock.”
The Nova Foundation Models take an interesting approach. Instead of just throwing out one model to rule them all, AWS built a whole suite designed for different needs:
Understanding models:
- Nova Micro - Text-only model optimized for speed and cost
- Nova Lite - Low-cost multimodal model for processing images, video, and text
- Nova Pro - Highly capable multimodal model with excellent accuracy-speed-cost balance
- Nova Premier - Their most capable model (coming soon)
Creative content generation:
- Nova Canvas - For image generation
- Nova Reel - For video creation
The Nova Differentiators:
- Different Context Windows: You can pick what works for your use case - no need to pay for a massive context window if you don't need it
- Performance Focus: They've put a lot of work into inference time optimization, which matters a lot in production
- Cost Efficiency: The smaller models can be surprisingly capable while being much more cost-effective
What's cool with Nova is that AWS isn't just playing the "bigger is better" game. They're acknowledging that different workloads need different tools. Need blazing-fast responses for a customer-facing app? There's a model for that. Working on more complex tasks that need deeper understanding? There's a model for that, too.
We are still waiting to see multimodal input to multimodal output, but that hasn’t come out yet. I’m patiently awaiting those features. That’ll be a huge advantage for a lot of use cases.
What's New with Amazon Nova
Just last week (March 19th), Amazon announced expanded Tool Choice parameter options for Nova's Converse API. This update enhances developers' control over model interactions with tools by supporting three different modes:
- Auto mode - Leaves tool selection entirely to Nova's discretion (the default option)
- Any mode - Prompts Nova to return at least one tool call from your specified list
- Tool mode - Enables developers to request a specific tool be returned by Nova
These options give developers much more granular control over how Nova interacts with their applications, particularly useful for machine-to-machine interactions and enforcing structured outputs.
Our Real-World Testing at Mission
My team has been rigorously experimenting with the Nova family of models as potential options for our customers. One area where we're seeing particularly promising results is using Nova as a potential replacement for other models in speed-critical applications.
Here's a real-world example: One of our customers is developing a next-generation call center IVR (Interactive Voice Response) system powered by generative AI. In this context, speed is absolutely critical - every millisecond of latency between a caller's query and the AI's response directly impacts the customer experience.
Nova Models are showing impressive performance in these low-latency scenarios. While other model offerings provide exceptional capabilities, Nova is purpose-built for speed with its 128K token context window and optimization for quick processing.
What's Next?
As we continue to test Nova, we're strategizing how to leverage Nova's capabilities to potentially reduce costs and latency for our customers across various use cases, from document processing to customer service automation.
Have you started experimenting with Amazon Nova yet? I'd love to hear about your experiences or thoughts on how you might implement these models in your workflows. Let me know what you think.
Until next time,
Ryan Ries
Now, here's our weekly AI-generated image & the prompt I used.
"Generate a painting inspired by Michelangelo's Sistine Chapel but use muppets as the subject of the painting."
Sign up for Ryan's weekly newsletter to get early access to his content every Wednesday.
