Викторина, связанная с руководством · Средний Уровень

Оценка затрат на LLM API и бюджет токенов. Викторина

Проверьте свои навыки прогнозирования расходов на LLM: количество токенов на запрос, соотношение входных и выходных цен, рост истории разговоров и ежемесячные бюджеты.

Связанные направляющие путиОценка стоимости LLM API
Предварительный просмотр каждого вопроса8 вопросы

Ответы и пояснения появляются по мере прохождения теста.

  1. In English, roughly how much text does one token represent on average?
    • About one full sentence
    • About four characters, or three quarters of a word
    • Exactly one word
    • About one paragraph
  2. Which type of token is usually priced higher by LLM API providers?
    • Input tokens
    • System prompt tokens
    • Cached tokens
    • Output tokens
  3. At a hypothetical price of $3 per million input tokens, what do 2,000 input tokens cost?
    • $0.006
    • $0.06
    • $0.0006
    • $6.00
  4. Why do chat conversations get more expensive as they get longer?
    • Providers charge a fee for each turn in a conversation
    • Output tokens get more expensive later in a conversation
    • The full conversation history is resent as input on every turn
    • Models slow down and bill by the second
  5. How are the internal thinking tokens of reasoning models typically billed?
    • They are free because users never see them
    • They are billed as output tokens even when not shown
    • They are billed as input tokens
    • They are billed at a flat daily rate
  6. Using the guide's worked example of $0.012 per request and 10,000 requests a day, what is the cost for a 30-day month before overhead?
    • $360
    • $1,200
    • $36,000
    • $3,600
  7. What is the effect of capping conversation history or summarizing older turns?
    • It flattens the growth of input tokens as conversations get longer
    • It raises the output price per token
    • It removes the need to count tokens
    • It turns output tokens into input tokens
  8. Besides cost, which API limit does the guide say can cap throughput?
    • A limit on the number of distinct users
    • A limit on the number of system prompts
    • Rate limits measured in tokens per minute
    • A limit on how many languages you can use