Token usage and cost tracking in Agentic AI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding Token Counting in Language Models

When using a language model, why is it important to count tokens instead of characters or words?

A Because words are always single tokens, so counting words is the same as counting tokens.

B Because characters are always longer than tokens, so counting characters is less efficient.

C Because tokens represent the smallest units the model processes, which may be parts of words, ensuring accurate cost and usage tracking.

D Because tokens are only used for billing, not for model input processing.

💡 Hint

Think about how language models break down text internally.

Language models split text into tokens, which can be whole words or parts of words. Counting tokens matches how the model processes input and output, which is essential for accurate cost and usage tracking.

❓ Predict Output

intermediate

Calculating Token Usage from Model Output

Given the following Python code that simulates token counting for a model response, what is the printed output?

Agentic AI

def count_tokens(text):
    # Simple token count: split by spaces
    return len(text.split())

input_text = "Hello world! This is a test."
output_text = "Hello! This test is simple."

input_tokens = count_tokens(input_text)
output_tokens = count_tokens(output_text)
total_tokens = input_tokens + output_tokens
print(total_tokens)

A 12

B 10

C 9

D 11

💡 Hint

Count the words in both input and output texts separately, then add.

The input text has 6 words, the output text has 5 words, so total tokens counted as words is 11.

❓ Hyperparameter

advanced

Choosing the Right Token Limit for Cost Efficiency

You have a language model with a maximum token limit of 4096 tokens per request. You want to maximize the amount of information processed while minimizing cost. Which strategy best balances token usage and cost?

A Set a token limit slightly below 4096 to leave room for model-generated tokens and avoid errors, optimizing cost and usage.

B Always use the full 4096 tokens to get the most output regardless of cost.

C Use only 512 tokens per request to save cost, even if it means sending multiple requests.

D Ignore token limits and send as much text as possible; the model will truncate automatically.

💡 Hint

Consider both model constraints and cost when setting token limits.

Setting a token limit slightly below the max allows input plus output tokens to fit without errors, optimizing usage and cost.

❓ Metrics

advanced

Calculating Cost from Token Usage

A language model charges $0.0004 per 1,000 tokens. If a user sends a prompt of 1,200 tokens and receives a response of 800 tokens, what is the total cost for this interaction?

A $0.0004

B $0.0008

C $0.00032

D $0.00048

💡 Hint

Add input and output tokens, then multiply by cost per 1,000 tokens.

Total tokens = 1,200 + 800 = 2,000 tokens. Cost = 2,000 / 1,000 * $0.0004 = $0.0008.

🔧 Debug

expert

Identifying Token Counting Bug in Code

Consider this Python function intended to count tokens by splitting text on spaces. It is used to track token usage for cost calculation. What error or issue will this code cause when processing the text "Hello,world!"?

def count_tokens(text):
    return len(text.split(' '))

print(count_tokens("Hello,world!"))

A It returns 1 because split(' ') does not split on multiple spaces correctly.

B It returns 2, correctly counting tokens separated by spaces.

C It raises a TypeError because split(' ') is invalid syntax.

D It returns 3 because punctuation is counted as separate tokens.

💡 Hint

Try running the split method on a string with punctuation and spaces.

Using split(' ') splits only on single spaces, so if the string has multiple spaces or tabs, it may not split correctly. For "Hello,world!" it returns 1 because the string has no exact single space but a comma and space combined.