AI Pricing Models: Per-Token, Cached, Batch, and Reserved Capacity

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-model-families-pricing-models-final5-creators

1. Which workload pattern is most economically suited for per-token pricing?

   1. A fixed daily workload of 10 million tokens
   2. Sporadic, low-volume requests that arrive unpredictably
   3. A nightly job processing millions of database records
   4. Continuous background data processing running 24/7

2. A developer implements prompt caching and sees costs drop by 60%. What scenario most likely explains this result?

   1. They switched from a premium model to a budget model
   2. Their system prompt and user queries share a long consistent prefix
   3. They reduced their maximum output token limit
   4. They began using a different API provider

3. A team runs the same complex analysis on 50,000 records every night. Which pricing approach would likely yield the greatest cost savings?

   1. Pay-per-request with priority routing
   2. Batch API with asynchronous processing
   3. Reserved capacity with dedicated infrastructure
   4. Per-token pricing with standard API calls

4. A company's AI usage is highly predictable: they process 2 million tokens daily, 365 days a year, with consistent request patterns. Which pricing tier best matches this situation?

   1. Reserved capacity or committed use discounts
   2. Pay-as-you-go with no commitments
   3. Per-token pricing with prompt caching enabled
   4. Batch API with background processing

5. A developer changes just one word in their system prompt and notices their cached tokens dropped to nearly zero. What does this demonstrate about prompt caching?

   1. The cache was deliberately cleared by the provider
   2. Cache hits require exact prefix matches
   3. Cached prompts have a 24-hour expiration policy
   4. Cache automatically rebuilds within a few minutes

6. What is a fundamental limitation that prevents AI systems from automatically selecting the optimal pricing tier?

   1. AI systems cannot connect to billing APIs
   2. AI cannot measure its own input and output token usage precisely
   3. The API does not provide usage data to the model
   4. Pricing tiers require human-level financial reasoning

7. A startup wants to minimize costs for an application where users submit questions and expect answers within 2 seconds. They have 10,000 daily users. Which approach would likely save them the most money?

   1. Switch to reserved capacity with per-second billing
   2. Move all processing to a serverless function
   3. Implement aggressive prompt caching on the system prompt
   4. Use batch API processing for all user queries

8. Which statement best describes the primary use case for cached tokens?

   1. Storing conversation history between sessions
   2. Lowering costs when the same long context appears repeatedly
   3. Enabling faster real-time speech translation
   4. Reducing costs for one-time, unique queries

9. An AI-powered customer service chatbot receives 100,000 messages per day with a 2,000-token system prompt that never changes. What cost optimization strategy should they prioritize?

   1. Switching to reserved capacity immediately
   2. Negotiating volume discounts with the API provider
   3. Maximizing prompt cache hits on the system prompt
   4. Implementing batch processing for all messages

10. A data science team needs to summarize 500,000 PDF documents over the weekend. They don't need results until Monday. Which pricing strategy would be most appropriate?

    1. Per-token pricing with standard synchronous API calls
    2. Batch API with asynchronous processing
    3. Reserved capacity with dedicated GPU instances
    4. Pay-per-request with premium support

11. What does the lesson identify as two specific moves that can cut AI bills by over 50%?

    1. Using smaller models and reducing output limits
    2. Enabling compression and caching all outputs
    3. Maximizing prompt cache hits and pushing async work to batch APIs
    4. Switching providers and negotiating contracts

12. Why might a company choose NOT to use reserved capacity pricing even with high, predictable volume?

    1. They need the flexibility to scale down during low periods without penalties
    2. The technology is not yet available for most AI providers
    3. Reserved capacity actually costs more than per-token for any volume
    4. Reserved capacity is only available to enterprise customers

13. Which scenario best represents an ideal use case for batch API pricing?

    1. A user waiting in a chat interface for an immediate response
    2. A content moderation system scanning millions of uploaded images overnight
    3. A stock trading algorithm making split-second decisions
    4. A healthcare system flagging urgent lab results in real-time

14. A team wants to reduce costs but finds their prompt changes frequently due to adding new features. What should they prioritize before investing in caching infrastructure?

    1. Stabilizing their prompts to ensure cache effectiveness
    2. Removing all system prompts to maximize cache hits
    3. Switching to a cheaper AI model
    4. Implementing a prompt versioning system

15. What is the primary trade-off when using batch API pricing?

    1. Lower cost in exchange for asynchronous (delayed) results
    2. Lower cost but must use more expensive premium models
    3. Higher cost but with unlimited rate limits
    4. Higher cost but guaranteed immediate results