CertifyCloudxAce every cloud certification.

AI-901 sample questions with answers

10 free practice questions for the Microsoft Azure AI Fundamentals (AI-901) exam. Try each one, then open the answer to see why the right option wins and every other option loses.

Question 1Identify AI concepts and capabilities

A bank runs a Foundry-based assistant for 4,000 staff. Usage is steady between 08:00 and 18:00, response latency must stay predictable at peak, and finance requires the monthly cost to be fixed in advance rather than varying with the number of tokens consumed. Which deployment option should you choose?

  1. A.

    A fine-tuned copy of the model published on a standard deployment

  2. B.

    A provisioned throughput deployment sized for the expected peak load

  3. C.

    A batch deployment that processes queued requests offline and returns results later

  4. D.

    A standard pay-as-you-go deployment that bills for each token consumed

Show answer

Answer: B

Provisioned throughput reserves capacity, which is what delivers predictable latency and a fixed monthly cost.

  • A. Fine-tuning changes model behaviour, not capacity allocation or billing model, so neither constraint is met.
  • B. Reserved capacity gives predictable throughput at peak and a committed price that does not vary with token volume.
  • C. Batch returns results asynchronously and cannot meet interactive latency expectations for 4,000 staff.
  • D. Per-token billing makes the monthly cost move with usage, which finance explicitly ruled out.
Question 2Identify AI concepts and capabilities

An insurer is building an application in which a claims adjuster uploads a photograph of vehicle damage and then asks questions about it in plain language, receiving written answers. You must pick a model from the Microsoft Foundry model catalog. Which type of model should you deploy?

  1. A.

    A text embedding model that converts text into numeric vectors

  2. B.

    A multimodal chat model that accepts image and text input and returns text

  3. C.

    A text-to-speech neural voice that reads text aloud

  4. D.

    An image generation model that creates images from a text description

  5. E.

    A speech recognition model that transcribes recorded audio

Show answer

Answer: B

The workload sends an image plus a question and expects text back, which only a multimodal chat model does.

  • A. Embedding models return vectors for similarity search and cannot produce a natural-language answer.
  • B. Image plus text in and text out is the signature of a multimodal chat completion model.
  • C. Text to speech only renders audio from text and cannot inspect an uploaded photograph.
  • D. Image generation runs the wrong direction: it creates a picture from text instead of interpreting one.
  • E. Speech recognition transcribes audio; there is no audio in this workload and no image understanding.
Question 3Identify AI concepts and capabilities

Lamna Healthcare is building a pipeline in which clinic notes are summarised by a deployed model. The notes contain patient names and record numbers, and the information governance lead has ruled that those identifiers must not leave the clinic's own systems. Clinic notes containing patient names and record numbers must be summarised by a model, but the identifiers must not leave the clinic's systems. What should the pipeline do first?

  1. A.

    Store the notes in the same resource group as the Foundry resource

  2. B.

    Raise the model deployment's temperature so that identifiers are paraphrased

  3. C.

    Detect and redact the personal identifiers in the notes before the text is sent to the model

  4. D.

    Switch the deployment to a model with a larger context window

  5. E.

    Increase the maximum output tokens so that the summary has room for context

Show answer

Answer: C

Personal identifiers must be removed before the text is sent, because nothing downstream can unsend them.

  • A. Resource group placement is an organisational convenience with no effect on prompt contents.
  • B. Temperature varies the wording of the completion and cannot prevent an identifier being transmitted.
  • C. Redacting identifiers before the request is the only measure that stops them being sent at all.
  • D. A larger context window lets the model read more text, which is the opposite of sending less.
  • E. A larger output allowance changes how long the summary may be, not what the prompt contains.
Question 4Identify AI concepts and capabilities

An application at Fabrikam Insurance lets a surveyor upload a photograph of a damaged roof and then asks the model to describe what the photograph contains. Which capability is being used?

  1. A.

    A translation model changing the language

  2. B.

    A multimodal model interpreting visual input

  3. C.

    A text to speech model producing audio

  4. D.

    An image generation model creating new visual output

  5. E.

    An embedding model producing a vector

Show answer

Answer: B

Describing the contents of an uploaded photograph is a multimodal model interpreting visual input.

  • A. A translation model changes the language of text and looks at no picture.
  • B. An image in and a description out is a multimodal model interpreting visual input.
  • C. Text to speech produces audio from written text and never examines an image.
  • D. Image generation runs the other way, producing a new picture from a written description.
  • E. An embedding model returns a vector for similarity comparison, not a description.
Question 5Identify AI concepts and capabilities

What is a token, as a large language model uses the term?

  1. A.

    A unit of the model's vocabulary, which can be a word, a sub-word or punctuation

  2. B.

    A numeric measure of how confident the model is in the answer it produced

  3. C.

    A single character of the input text, stored with its position in the sequence

  4. D.

    A fixed-length vector that encodes the meaning of a complete sentence

  5. E.

    A credential presented to the endpoint so that the request is authorised

Show answer

Answer: A

A token is a vocabulary unit, which may be a whole word, a sub-word, punctuation or another common character sequence.

  • A. Tokens are vocabulary units covering words, sub-words, punctuation and common character sequences.
  • B. Confidence scores are a separate output and are not what the word token denotes.
  • C. Models tokenise into units larger than single characters, and position is recorded separately.
  • D. A vector encoding meaning is an embedding, which is produced from tokens rather than being one.
  • E. That is an access token, an unrelated use of the same word for an authorisation credential.
Question 6Identify AI concepts and capabilities

A publisher must classify two million archived articles by subject. The work is scheduled to run overnight and nobody is waiting for any individual result. A publisher must classify two million archived articles overnight and does not need results in real time. Which deployment type fits?

  1. A.

    Global Batch

  2. B.

    Global Standard, which offers the highest default quota for interactive pay-per-token traffic

  3. C.

    Instant access, which allows a supported model to be called without creating a deployment

  4. D.

    Regional Provisioned, which reserves capacity within one customer-specified Azure geography

  5. E.

    Developer, which provides a temporary endpoint for evaluating a fine-tuned model

Show answer

Answer: A

Large asynchronous jobs that can wait belong on a batch deployment, at half the price with a twenty-four hour target.

  • A. Batch handles large asynchronous volumes at half the price with a twenty-four hour target turnaround.
  • B. Global Standard serves interactive traffic and would cost twice as much for this work.
  • C. Instant access removes the deployment step but does not provide batch pricing or queueing.
  • D. Regional Provisioned reserves capacity in one geography, which the scenario does not require.
  • E. Developer is for evaluating a fine-tuned model and expires after twenty-four hours.
Question 7Identify AI concepts and capabilities

A team must decide which review comments express dissatisfaction. Which AI workload is this?

  1. A.

    Computer vision

  2. B.

    Text analysis

  3. C.

    Speech

  4. D.

    Information extraction from video

  5. E.

    Image generation

Show answer

Answer: B

Judging whether written comments express dissatisfaction is sentiment analysis, a text analysis workload.

  • A. Computer vision workloads take images or video as input, which this scenario does not have.
  • B. Judging the attitude expressed in written text is sentiment analysis, a text analysis workload.
  • C. Speech workloads involve audio in or out, and this scenario has only written comments.
  • D. Information extraction from video applies to recordings, not to written review comments.
  • E. Image generation produces pictures from descriptions rather than judging existing text.
Question 8Identify AI concepts and capabilities

A manufacturer is planning six features for its production systems and must identify which of them need a model that produces new visual output rather than analysing an existing image. Which two require a model that produces new visual output? Each correct answer presents a complete solution. (Choose TWO.)

Choose 2.

  1. A.

    Create a concept rendering of a new casing from a written specification

  2. B.

    Read the batch code printed on each component in a photograph

  3. C.

    Generate a replacement background for an existing product photograph

  4. D.

    Transcribe the line supervisor's spoken shift handover

  5. E.

    Describe what is happening in a photograph of an assembly line

  6. F.

    Count the finished units visible on a pallet in a warehouse photograph

Show answer

Answer: A, C

Creating a concept rendering and generating a replacement background both produce new visual output.

  • A. A concept rendering created from a written specification is new visual output from a prompt.
  • B. Reading a printed batch code is optical character recognition, which returns text.
  • C. Replacing a background in an existing photograph is image editing, which produces new visual output.
  • D. Transcribing a spoken handover is speech to text, whose input is audio.
  • E. Describing an assembly line photograph is interpretation of visual input, returning text.
  • F. Counting units on a pallet analyses an existing image and returns a number.
Question 9Identify AI concepts and capabilities

An organisation has approved only two deployment types for its Foundry projects and must stop teams from creating any others, rather than discovering unapproved deployments in a later review. Which mechanism enforces this?

  1. A.

    Azure Policy

  2. B.

    A written standard circulated to all engineering teams at the start of each quarter

  3. C.

    A guardrail with the hate and violence content controls set to a high severity threshold

  4. D.

    A monthly report listing the deployments each team created during the preceding period

  5. E.

    A quota increase request raised with support before any new deployment is created

Show answer

Answer: A

Azure Policy enforces organisational standards and can deny creation of a deployment with a disallowed SKU.

  • A. Azure Policy evaluates resource properties such as the deployment SKU and can deny non-compliant creation.
  • B. A circulated standard is guidance and is not enforced by the platform at creation time.
  • C. A content guardrail governs which completions are flagged, not which deployment types may exist.
  • D. A monthly report detects breaches after they happen rather than preventing them.
  • E. A quota request concerns throughput capacity, not which deployment types are permitted.
Question 10Identify AI concepts and capabilities

Which principle asks whether an AI solution empowers everyone and engages people regardless of ability, language or circumstance?

  1. A.

    Accountability

  2. B.

    Transparency

  3. C.

    Fairness

  4. D.

    Reliability and safety

  5. E.

    Inclusiveness

Show answer

Answer: E

Empowering everyone and engaging people regardless of ability, language or circumstance is the inclusiveness principle.

  • A. Accountability concerns named human owners who retain meaningful control over the system.
  • B. Transparency concerns disclosing how the system behaves and where its limits lie.
  • C. Fairness compares outcomes for similar people; it does not ask who can use the system at all.
  • D. Reliability and safety concerns correct operation and safe behaviour under unexpected conditions.
  • E. Empowering everyone regardless of ability, language or circumstance is Microsoft's inclusiveness wording.

Keep going with 490 more AI-901 questions

Free papers every day, in the real exam formats, with progress by exam domain. Unlock every paper and timed mock exam when you are ready.

Practise freeSee all AI-901 papers
AI-901 sample questions with answers (10 free) · CertifyCloudx