AI Prompt Caching: 90% Discount on Repeated Context

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-tools-ai-prompt-caching-r13a2-creators

1. What is the primary financial advantage of using prompt caching with providers like Anthropic and OpenAI?

   1. Cached content is processed faster, reducing compute fees
   2. Cached tokens receive up to a 90% discount compared to regular token pricing
   3. Caching reduces the total number of tokens sent in each request
   4. Cached prompts eliminate all API costs for repeated requests

2. Within what time frame does the cached version of a system prompt remain valid for reuse?

   1. Exactly 1 hour
   2. Approximately 5 minutes
   3. Only during the current API session
   4. Until the user starts a new conversation

3. Where should you place stable content like system prompts and large reference documents in your API calls to maximize caching benefits?

   1. Scattered throughout the prompt for better organization
   2. At the end of the prompt (after user message)
   3. At the beginning of the prompt (first position)
   4. It doesn't matter where content is placed

4. What happens to your caching savings if you make even a small edit to your system prompt between requests?

   1. The cache is invalidated and you pay full price for the next request
   2. The system automatically merges the old and new cache
   3. The cache remains active but charges a small editing fee
   4. The edited prompt creates a new cache entry without affecting the old one

5. How does prompt caching typically affect the latency of subsequent API calls with identical prefixes?

   1. Latency decreases because the cached prefix doesn't need reprocessing
   2. Latency stays the same but costs decrease
   3. Latency increases due to cache lookup overhead
   4. Latency becomes variable and unpredictable

6. Which type of content is most effective to cache for cost savings?

   1. User messages that vary with each query
   2. Stable system instructions and large documents that don't change between requests
   3. Real-time data feeds
   4. Temporary session variables

7. In a RAG (Retrieval-Augmented Generation) setup, what type of content benefits most from prompt caching?

   1. The user's unique question in each request
   2. Temporary authentication tokens
   3. Stable retrieved chunks that are consistent across multiple queries
   4. Dynamic search results that change each time

8. What does it mean that cache invalidates if any 'prefix byte changes'?

   1. Only changes to the last byte of the prompt affect caching
   2. Changing even one byte at the start of your prompt breaks the cache entirely
   3. The cache system rounds up byte counts to the nearest kilobyte
   4. Unicode characters invalidate the cache but ASCII does not

9. Which combination of techniques would likely yield the greatest cost savings?

   1. Prompt caching combined with batch processing APIs
   2. Requesting real-time data updates with each call
   3. Using multiple different system prompts
   4. Increasing the number of tokens in each request

10. Why is versioning your system prompts important when using prompt caching?

    1. Versioning automatically optimizes cache efficiency
    2. Versioning has no impact on caching behavior
    3. Different versions create separate cache entries, allowing you to roll back without losing savings
    4. You must version prompts because caching requires unique identifiers

11. What would be the worst approach for maximizing prompt caching savings?

    1. Using the same cache breakpoint markers consistently
    2. Keeping system prompts consistent across requests
    3. Placing user questions before system prompts
    4. Caching large reference documents

12. What is the relationship between cache breakpoints and provider documentation?

    1. Cache breakpoints are automatically detected by all AI providers
    2. There is no such thing as cache breakpoints in prompt caching
    3. Cache breakpoints must be explicitly marked according to each provider's documentation
    4. Cache breakpoints are only relevant for free-tier users

13. What occurs when the provider-defined TTL (time-to-live) for a cached prompt expires?

    1. The cached content must be reprocessed on the next request, at full price
    2. The user receives a notification that caching has expired
    3. The system deletes the prompt and sends an error notification
    4. The cache automatically moves to long-term storage

14. A developer sends the same system prompt followed by a different user question every minute for 10 minutes. How many times will the system prompt portion be charged at full price?

    1. Only the first time
    2. Once at the beginning and once at the end
    3. Never, because questions are different
    4. Every time because questions differ

15. What does the '90% discount' specifically apply to in prompt caching?

    1. The number of requests you can make
    2. The total API bill including non-cached tokens
    3. Only the cached tokens themselves, not the entire request cost
    4. The base subscription fee