Latency and cost benchmarking in Agentic AI - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Metrics

intermediate

Understanding latency measurement units

You run a latency benchmark on an AI model and get a result of 250 ms. What does this number represent?

A The size of the model in megabytes

B The total cost in dollars to run the model for one hour

C The number of inputs the model can process per second

D The time it takes for the model to process one input from start to finish

💡 Hint

Latency is about time, not cost or size.

Latency measures how long the model takes to respond to a single input, usually in milliseconds.

❓ Model Choice

intermediate

Choosing a model for low cost and moderate latency

You want to deploy an AI model that balances low cost and moderate latency for a chatbot. Which model type is best?

A A very large transformer model with billions of parameters

B A small transformer model optimized for fast inference

C A rule-based system with no machine learning

D A deep convolutional neural network designed for image tasks

💡 Hint

Smaller models usually cost less and run faster.

Small transformer models are designed to run efficiently with reasonable latency and cost, suitable for chatbots.

❓ Predict Output

advanced

Calculating average latency from benchmark data

What is the output of this Python code that calculates average latency in milliseconds?

Agentic AI

latencies = [120, 150, 130, 160, 140]
avg_latency = sum(latencies) / len(latencies)
print(f"Average latency: {avg_latency} ms")

A Average latency: 700 ms

B Average latency: 150 ms

C Average latency: 140.0 ms

D SyntaxError

💡 Hint

Sum all values and divide by count.

The sum is 700 and there are 5 values, so average is 700/5 = 140.0 ms.

🔧 Debug

advanced

Identifying the cause of high cost in benchmarking

You benchmarked two AI models and found one costs 10x more to run despite similar latency. What is the most likely cause?

A The expensive model uses more compute resources per request

B The cheaper model has higher latency

C The benchmarking code has a syntax error

D The expensive model is smaller in size

💡 Hint

Cost depends on compute usage, not just latency.

Higher cost usually comes from more compute power or memory used per request, even if latency is similar.

🧠 Conceptual

expert

Interpreting latency and cost trade-offs in AI deployment

Which statement best explains why reducing latency might increase cost in AI model deployment?

A Using more powerful hardware to reduce latency usually increases operational cost

B Reducing latency always reduces cost because the model runs faster

C Latency and cost are unrelated metrics in AI deployment

D Increasing latency reduces cost because it uses more resources

💡 Hint

Think about hardware and resource usage.

Faster hardware or more parallel processing reduces latency but costs more to operate.