Answer: On-Demand

## Analysis of Amazon Bedrock Pricing Models Based on the question requirements, the company needs: 1. **Limited budget** - Cost control is essential 2. **Flexibility** - Ability to scale usage up or down 3. **No long-term commitment** - Avoid contracts or reservations ### Evaluation of Options: **A: On-Demand** - **CORRECT** - **Pay-as-you-go model**: Charges only for actual usage (input tokens processed and output tokens generated) - **No upfront costs or commitments**: No minimum fees or long-term contracts required - **Maximum flexibility**: Can start, stop, or adjust usage at any time without penalties - **Ideal for variable workloads**: Perfect for development, testing, or applications with unpredictable usage patterns - **Budget control**: Costs scale directly with usage, allowing precise budget management **B: Model customization** - **INCORRECT** - This refers to fine-tuning or customizing foundation models, not a pricing model - Customization typically involves additional costs and may require more stable usage patterns - Does not address the core requirement of flexible pricing without commitment **C: Provisioned Throughput** - **INCORRECT** - Requires **commitment to a specific capacity level** for a minimum duration (typically 1-6 months) - Involves **reserved capacity payments** regardless of actual usage - Better suited for **predictable, steady-state production workloads** with consistent demand - Contradicts the requirement for "no long-term commitment" **D: Spot Instance** - **INCORRECT** - **Not applicable to Amazon Bedrock** - Spot Instances are an EC2 pricing model for compute capacity - Amazon Bedrock is a managed service with different pricing constructs - Even if conceptually similar, Bedrock doesn't offer spot pricing for inference ### Why On-Demand is Optimal: The On-Demand pricing model provides the perfect balance of cost control and flexibility for companies with limited budgets and uncertain usage patterns. It eliminates financial risk by avoiding upfront commitments while allowing the company to pay only for what they actually use. This model is particularly well-suited for: - Initial development phases where usage patterns are unpredictable - Applications with variable or seasonal demand - Companies testing AI capabilities before committing to larger investments - Projects where budget constraints require precise cost tracking Other models either don't exist for Bedrock (Spot) or require commitments that contradict the stated requirements (Provisioned Throughput). Model customization is a feature, not a pricing model.

Author: LeetQuiz Editorial Team