Nova Pro (non-reasoning)

Given Nova Pro's competitive input token pricing, keeping prompts concise yet clear is still a best practice. While the cost per token is low, unnecessary verbosity in prompts can still add up over millions of calls.

• **Be Direct:** Formulate prompts that get straight to the point, providing all necessary context without extraneous details.
• **Leverage Examples:** Instead of lengthy instructions, use a few clear in-context examples to guide the model's output format and style.
• **Pre-process Inputs:** Clean and condense input data before sending it to the model to reduce token count.

Nova Pro is noted for its concise outputs, which is a significant cost advantage. Actively encourage this behavior in your prompts.

• **Specify Length:** Include explicit instructions like "Summarize in 3 sentences" or "Provide a brief description."
• **Define Output Format:** Use structured formats (e.g., bullet points, JSON) to guide the model to produce only essential information.
• **Post-process Outputs:** Implement a post-processing step to trim any unnecessary introductory or concluding remarks the model might generate.

For non-real-time applications, consider batching multiple requests into a single API call if the Bedrock API supports it for Nova Pro, or process multiple items concurrently.

• **Group Similar Tasks:** Combine multiple small summarization or generation tasks into a larger batch to potentially reduce overhead.
• **Asynchronous Processing:** Utilize asynchronous calls for tasks that don't require immediate responses, allowing for more efficient resource utilization.

The 300k context window is powerful, but using it judiciously is key to cost control. Only include necessary information.

• **Dynamic Context:** Implement logic to dynamically include only the most relevant parts of a conversation history or document for each query.
• **Summarize History:** For very long interactions, periodically summarize past turns to keep the context window manageable without losing critical information.

Regularly track your token usage and associated costs on Amazon Bedrock to identify trends and potential areas for optimization.

• **Set Up Alerts:** Configure AWS cost explorer alerts for Nova Pro usage to be notified of unexpected spikes.
• **Analyze Usage Patterns:** Understand which applications or features are consuming the most tokens and focus optimization efforts there.

Nova Pro excels in cost-efficiency, multimodal input (text and image), and generating concise text outputs. It's ideal for high-volume content generation, summarization, and image captioning tasks where complex reasoning is not the main requirement.

Nova Pro scores 25 on the Artificial Analysis Intelligence Index, placing it below the average of 30 for comparable models. This indicates it is less suited for complex reasoning or highly nuanced tasks, but its pricing makes it very competitive for its intended use cases.

Nova Pro features a substantial 300k token context window, allowing it to process and generate longer, more coherent pieces of text or analyze larger volumes of input data effectively.

Yes, Nova Pro supports multimodal input, meaning it can process both text and image inputs. Its output, however, is text-only.

On Amazon Bedrock, Nova Pro is priced at $0.80 per 1M input tokens and $3.20 per 1M output tokens. Its blended price (based on a 3:1 input:output ratio) is $1.40 per 1M tokens, making it highly competitive.

It's best suited for tasks requiring efficient, high-volume text generation, summarization, data extraction, and image captioning. Examples include generating product descriptions, summarizing long documents, or creating social media content.

Nova Pro has a median output speed of 78 tokens per second and a low latency (Time to First Token) of 0.42 seconds on Amazon Bedrock, ensuring good performance for most applications.