Help us improve the AWS re:Post Knowledge Center by sharing your feedback in a brief survey. Your input can influence how we create and update our content to better support your AWS journey.
Amazon Bedrock Pricing Questions
I have questions regarding certain pricing aspects of using the Bedrock service for machine learning fine-tuning of the Llama 3 70B model on us-west-2 (this is the closest region to me that supports Llama 3).
Firstly, as seen in this image: 
Secondly, regarding pricing optimisation etc., which would be the better option to go for - Amazon Bedrock Titan Lite or Amazon Bedrock Titan Express? My use-case involves training the model on about 2 million tokens (includes inputs/outputs) of character text. One input may be about a couple hundred words in length mapping onto one output of slightly less length. So unless there's reason to go for Amazon Bedrock Titan Express, would it not be better to just go for Titan Lite as it's far cheaper?
Thirdly, when purchasing a provisioned throughput on no-commit, is it absolutely guaranteed that you will only be billed for the computing power you used? Since it's also an hourly cost, would one single prompt that may take seconds/minutes to return an answer still be considered use of the full hour or would it add up over time? (i.e. if you used 5 minutes every day for 12 days that would only be considered 1 hour of usage and as such be billed like this)
Thank you very much for your time.
- Newest
- Most votes
- Most comments
Hi again! I hope I can clarify your questions. I'll go through each item, but feel free to ask if anything remains unclear.
1. Monthly Cost for Fine-Tuning
The monthly cost reflects the ongoing use of compute resources. According to the Amazon Bedrock pricing page, charges are based on the total tokens processed during training across all epochs, making it a recurring fee rather than a one-time cost.
2. Titan Lite vs. Titan Express
Given your use case, Titan Lite should be more cost-effective unless you need the higher throughput of Titan Express. Titan Lite is well-suited for smaller datasets.
3. Provisioned Throughput Billing
Provisioned throughput is typically purchased on a monthly basis, meaning you commit to a certain level of compute power for the entire month, regardless of how much you actually use it. This is different from the hourly billing for on-demand usage.
For more details, refer to the Amazon Bedrock pricing.
Relevant content
- asked a year ago
Hi, re. fine-tuning monthly cost, look at "Model customization" on https://aws.amazon.com/bedrock/pricing/, it says "for model storage charged per month per model."