How to Lower Amazon Bedrock Monthly Cost with Few-Shot Prompting?
A company is using few-shot prompting on a base model that is hosted on Amazon Bedrock. The model currently uses 10 examples in the prompt. The model is invoked once daily and is performing well. The company wants to lower the monthly cost. Which solution will meet these requirements?
Community Votes
100% of anonymous learners picked answer B. Votes are pick records left by other test-takers — they are not the verified answer.
Community Insight
This question tests the understanding that Bedrock's on-demand pricing is token-based, so shrinking the input prompt is the most straightforward way to cut costs when invocation frequency is already low.
Amazon Bedrock charges based on the total number of input and output tokens processed per invocation. Reducing the number of tokens in a few-shot prompt directly lowers the per-invocation cost without requiring architectural changes.
Candidates often choose Provisioned Throughput (D) thinking it is cheaper, but PT requires a fixed monthly commitment and only saves money at high, predictable invocation volumes — the opposite of a once-daily use case.
Community Discussion (3 comments)
Comments & Corrections
No comments yet — spotted an error or have a note? Share it below.
Expert Analysis
Understanding Amazon Bedrock Pricing
Amazon Bedrock offers two primary pricing models:
- On-Demand: You pay per input token and output token processed during each API call. There is no monthly commitment.
- Provisioned Throughput: You reserve a dedicated block of model capacity for a fixed monthly or hourly fee, which is cost-effective only when invocation volume is high and predictable.
Why Option B is Correct
The scenario states the model is invoked only once daily and is already performing well with 10 few-shot examples. Because the company is on the on-demand pricing model, every token in the prompt — including the 10 examples — is billed on each invocation. Decreasing the number of tokens in the prompt (for example, by trimming examples, shortening instructions, or using more concise phrasing) directly reduces the input token count, which lowers the cost per call and, consequently, the monthly bill.
Why the Other Options Are Wrong
- A. Fine-tuning: Fine-tuning creates a customized model and incurs training costs plus ongoing inference costs. It does not inherently reduce prompt token counts and would add expense for a once-daily workload.
- C. Increase the number of tokens: Adding more tokens would increase the input token count and raise the cost per invocation — the exact opposite of the goal.
- D. Provisioned Throughput: As community member AzureDP900 incorrectly suggested, PT is not pay-per-use. It requires a committed monthly fee regardless of how often the model is called. For a model invoked once per day, PT would be dramatically more expensive than on-demand pricing.
Key Takeaway
When invocation frequency is low, always optimize the prompt size first. Fewer input tokens = lower cost per call = lower monthly bill on Bedrock's on-demand tier.
Official Reference
Exam Strategy
When a question mentions low invocation frequency and cost reduction on a token-billed service, immediately look for the option that reduces per-invocation resource consumption rather than committing to reserved capacity.
Related Analysis
Practice All AIF-C01 Questions
Access 100 questions with complete answers and detailed explanations.
View Full AIF-C01 Practice Test →