Free sampler

Try 12 free Microsoft AI-901 practice questions

Real exam-style questions taken word for word from the Microsoft AI-901 study guide. Three are below, with the full answer explanation. Enter your email and the full 12-question PDF is yours.

Unlock the full sampler
Microsoft AI-901 cover
12 questions in the PDF Taken verbatim from the book One email, no spam, unsubscribe any time

Three questions to try right now

Work each one before you open the answer. These are the same style and depth as the Azure AI Fundamentals (AI-901) exam expects.

Question 1. An engineering team must select a model to solve a multi-step mathematical optimization problem where an incorrect intermediate step invalidates the final answer. Cost and latency are secondary concerns. Which model type is MOST appropriate?

  • A. A small language model (SLM) for its low latency
  • B. A standard large language model (LLM) for its broad knowledge
  • C. A reasoning model trained to produce explicit step-by-step logic chains
  • D. An embedding model to find similar solved problems in a knowledge base
Show the answer

Answer: C. Reasoning models are explicitly designed for tasks that require accurate multi-step logic, mathematical derivations, or structured problem-solving. They produce explicit intermediate reasoning steps, which reduces compounding errors. The scenario states that an error at any intermediate step invalidates the result and that cost and latency are not constraints, the conditions that favor a reasoning model over a standard LLM.

  • A: SLMs prioritize speed and cost for simple tasks. A complex multi-step optimization problem is the opposite of the scenario SLMs are optimized for.

Question 2. A deployed chat model in Foundry is returning HTTP 429 errors during peak hours but performs normally during off-peak hours. The deployment is Standard type. Which setting is MOST likely causing the throttling, and what is the correct fix?

  • A. The temperature is too high; lower the temperature to reduce response variability
  • B. The Tokens-per-Minute quota is too low for peak demand; increase the TPM limit up to the account's regional ceiling
  • C. The model version is outdated; upgrade to the latest version to improve throughput
  • D. The content filter sensitivity is too strict; relax the filter to allow more requests through
Show the answer

Answer: B. HTTP 429 (rate limit exceeded) errors on a Standard deployment are caused by the Tokens-per-Minute quota ceiling being hit. TPM is a rate limit that caps how many tokens the deployment can process per minute. During peak hours, if the application's request rate exceeds the configured TPM, subsequent requests are throttled until the next minute window. The fix is to increase TPM up to the subscription's regional quota ceiling.

  • A: Temperature controls output randomness and has no effect on request throughput or rate limiting. Changing temperature will not resolve 429 errors.

Question 3. A business analyst asks why the organization spent money on a Provisioned throughput deployment instead of a Standard deployment. The engineering lead explains the key benefit. Which benefit is MOST accurate?

  • A. Provisioned throughput eliminates all content filtering, improving response speed
  • B. Provisioned throughput provides predictable latency and guaranteed throughput at a fixed cost, suitable for production workloads with known steady traffic
  • C. Provisioned throughput automatically scales to unlimited capacity with no quota ceiling
  • D. Provisioned throughput reduces the number of tokens billed per request by compressing the prompt
Show the answer

Answer: B. Provisioned throughput (PTU) reserves dedicated compute in Azure for the specific deployment. This eliminates the variability associated with shared capacity in Standard deployments, resulting in predictable latency and a guaranteed throughput ceiling. The trade-off is a fixed cost commitment regardless of actual usage. It is appropriate when the workload has known steady traffic and consistent latency is a service-level requirement.

  • A: Content filtering is a safety guardrail configured independently of deployment type. Provisioned deployments still apply content filters. Response speed improvement in PTU comes from dedicated compute, not from bypassing safety checks.

Unlock the full 12-question sampler

All 12 come with the same written explanations the book uses, covering why the right answer is right. We email you the PDF, nothing else.

One email with the PDF. Unsubscribe any time.

What you get

The sampler is free. Here is exactly what lands in your inbox.

The 12-question PDF

Print it or work it on screen, then score yourself against the answer key.

Questions from the real book

Nothing is written for marketing. Every question is lifted from the study guide itself.

A one-time discount

A few days later we send one email with a code for the full study guide. That is the entire sequence.

Microsoft AI-901 cover

The full Microsoft AI-901 study guide

The complete guide covers the Azure AI Fundamentals (AI-901) exam objective by objective in plain English, with the full practice question bank behind these samples.

From $20.99

See the study guide   All Microsoft guides