CBSE Class 10 · SQP 2026-27 · Last verified 3 November 2026

CBSE Class 10 AI (417) sample paper 2026-27, solved and explained

In short

This walkthrough covers all 21 questions of CBSE’s 2026-27 sample question paper for Class 10 Artificial Intelligence (417), using CBSE’s own question numbers. For each one we say in our own words what it asks, give the answer and show the reasoning or the working. Keep CBSE’s paper open alongside: its diagrams and exact wording are in the PDF we link, and its marking scheme shows how marks are split.

  • The paper is 50 marks in 2 hours: 21 questions, of which a student answers 15.
  • Section A (24 marks) has five questions of six one-mark parts each, with a choice inside every question.
  • Section B (26 marks) has 16 questions; students answer 10, in 20–30 words for 2-mark questions and 50–80 words for 4-mark ones.
  • The numericals, Q4 (vi), Q5 (iv), Q5 (vi) and Q20, are dependable marks once the formulas are learned.
  • Two answer lines in CBSE’s marking scheme do not match their questions: Q2 (iv) and Q15. We explain both below.

How is the 2026-27 Class 10 AI paper structured?

Sections, choice and marks
QuestionsWhat they testAttemptMarks
Q1Employability Skills, objective4 of 6 parts4
Q2–Q5Subject units, objective5 of 6 parts in each20
Q6–Q10Employability Skills, short answer3 of 56
Q11–Q16Subject units, short answer4 of 68
Q17–Q21Subject units, long answer3 of 512
21 questions15 answered50

From: Artificial Intelligence (Subject Code 417): Sample Question Paper and Blue-print, Class X, blue-print and general instructions, pp. 1–2

Use the paper marked “Session 2026-2027”. CBSE’s skill education portal hosts it with its marking scheme; on 3 November 2026 the older skill-subject page on cbseacademic.nic.in still linked last year’s files. The theory paper tests Units 1, 2, 3, 5 and 6 only; Units 4 and 7 are assessed in practicals. See our units and marks page for the full breakdown.

From: Artificial Intelligence (Subject Code 417): Sample Question Paper and Blue-print, Class X, header and blue-print; Artificial Intelligence (Sub. Code 417), Class X: Curriculum, scheme of studies, p. 3

Q1: Employability Skills objective questions (answer any 4)

Q1 (i)1 markMCQCommunication Skills-II

Which of the four options is not part of the communication process cycle?

Answer

(d) Time.

  • The cycle is made of a sender, a message, the channel it travels through, a receiver and the feedback that comes back. Time is not one of its elements.
Q1 (ii)1 markSequencing MCQSelf-Management Skills-II

Three steps for handling stress are listed out of order: using a method to manage it (i), noticing that you are stressed (ii) and finding its cause (iii). Which order is right?

Answer

(d) ii → iii → i.

  • You can only deal with stress once you notice it, and you can only pick a sensible method once you know what is causing it.
Q1 (iii)1 markMCQSelf-Management Skills-II

Ravi works hard all year because he wants the school’s best-student prize. Which type of motivation drives him?

Answer

(b) External.

  • The driving force is a reward from outside, the award, so this is external (extrinsic) motivation. Working hard because he enjoys the subject would be internal motivation.
Q1 (iv)1 markMCQICT Skills-II

Which file extension does a Notepad text file use?

Answer

(d) .txt.

  • Notepad saves plain text as .txt. A .jpg file is an image, .doc is a word-processor document, and “.text” is not the standard extension.
Q1 (v)1 markMCQEntrepreneurial Skills-II

Sanchi has just started a business. Which of the options is not one of an entrepreneur’s functions?

Answer

(c) Blaming employees for wrong decisions.

  • Taking risks and managing the business, creating new ideas or products, and dividing the income are all entrepreneur functions. Blaming staff is not.
Q1 (vi)1 markMCQGreen Skills-II

Narayana Peesapaty makes spoons from grain that can be eaten, and that rot harmlessly in soil if thrown away. Which Sustainable Development Goal does this serve?

Answer

(a) Responsible Consumption and Production (SDG 12).

  • The product replaces throwaway cutlery and leaves no lasting waste, which is about how goods are produced and used.

Q2: subject objective questions, set 1 (answer any 5)

Q2 (i)1 markMCQ with imageUnit 6: Natural Language Processing

A picture shows someone saying “Relax! I’m just pulling your leg” and two readings of it: that the speaker is joking, and that no leg is actually being pulled. Which stage of NLP does this show?

Answer

(d) Pragmatic analysis.

  • Working out what a speaker means in context, rather than what the words literally say, is pragmatic analysis. Syntactic analysis checks grammar, and discourse integration links a sentence to the sentences around it.
Q2 (ii)1 markAssertion–reasonUnit 5: Computer Vision

The assertion says computer vision lets machines make sense of images and videos. The reason says computer vision works by using sensors to feel the texture of objects.

Answer

(c) The assertion is true but the reason is false.

  • Computer vision works from visual input such as camera images. Sensing texture by touch is not how it works, so the reason is false.
Q2 (iii)1 markMCQUnit 1: AI Project Cycle

What is the model-building stage of the AI project cycle mainly for?

Answer

(c) Training and testing AI algorithms.

  • Building the model means choosing an algorithm, training it on the data and checking how it performs. Drawing graphs of the data belongs to data exploration.
Q2 (iv)1 markMCQUnit 3: Evaluating Models

What does a high F1 score usually suggest about a model?

Answer

(b) Precision and recall are well balanced.

  • F1 is the harmonic mean of precision (P) and recall (R): F1 = 2PR / (P + R). It is high only when both P and R are high, so a high F1 means the two are in good balance.

Note: CBSE’s marking scheme prints “c) Evaluating a model to determine its performance” against this question, which is not one of the four options in the paper. Read as printed, the answer is (b).

Q2 (v)1 markMCQ with imageUnit 2: Advanced Concepts of Modeling in AI

A picture shows labelled photos of apples and strawberries going into a machine learning model, which then labels a new fruit “Apple”. What type of learning model is this?

Answer

(a) Classification.

  • The model learns from labelled examples and predicts a category for a new item: supervised classification. Regression would predict a number, and clustering and association work on unlabelled data.
Q2 (vi)1 markCase-based MCQUnit 3: Evaluating Models

A flood-warning model alerts nearby villages to a flood, but no flood happens. What is this outcome called?

Answer

(c) False positive.

  • The model predicted the positive case (flood) but reality was negative (no flood), which is a false positive. Missing a real flood would be a false negative.

Q3: subject objective questions, set 2 (answer any 5)

Q3 (i)1 markMCQUnit 6: Natural Language Processing

In which of the four sentences does the word “bark” mean the outer covering of a tree?

Answer

(b) The sentence describing a tree’s bark as rough and peeling.

  • In the other sentences, bark is a dog’s sound or a way of shouting orders. Picking the right meaning of a word from its context is a semantic problem for NLP.
Q3 (ii)1 markMCQUnit 5: Computer Vision

What does object detection mean in computer vision?

Answer

(b) Finding objects in images or videos.

  • Object detection locates each object in an image or video frame and labels it, usually with a box drawn around it.
Q3 (iii)1 markMCQUnit 1: Ethical Frameworks for AI

What does non-maleficence, one of the principles of bioethics, mean?

Answer

(b) Do no harm.

  • It is one of the four principles of bioethics, alongside respect for autonomy, beneficence and justice.
Q3 (iv)1 markTwo statementsUnit 1: Ethical Frameworks for AI

Statement 1 says value-based frameworks keep AI in line with values people care about, like fairness, privacy and respect. Statement 2 says these frameworks tell developers to disregard social and cultural differences.

Answer

(c) Only Statement 1 is correct.

  • Value-based frameworks put human values at the centre, so they ask developers to take diversity into account, not to ignore it.
Q3 (v)1 markMCQUnit 2: Advanced Concepts of Modeling in AI

A video-game AI gets better by playing against itself, earning rewards for wins and penalties for losses. What type of learning is it?

Answer

(c) Reinforcement learning.

  • Learning by trial and error from rewards and penalties is reinforcement learning. There are no labelled examples, so it is not supervised learning.
Q3 (vi)1 markMCQUnit 3: Evaluating Models

Which option correctly describes precision?

Answer

(b) Correct positive predictions divided by all positive predictions.

  • Precision = TP / (TP + FP): of everything the model labelled positive, the share that really was positive. Option (a) describes recall and option (c) describes accuracy.

Q4: subject objective questions, set 3 (answer any 5)

Q4 (i)1 markMCQUnit 5: Computer Vision

How is computer vision used in online shopping?

Answer

(c) Visual search and virtual try-on.

  • Shoppers can search with a photo, or see clothes and glasses on themselves through the camera. Both depend on the computer understanding images.
Q4 (ii)1 markMCQUnit 6: Natural Language Processing

What is the lemma of the word “happier”?

Answer

(d) happy.

  • Lemmatization reduces a word to its dictionary form, so “happier” becomes “happy”. A lemma is always a real word, which a stem need not be.
Q4 (iii)1 markMCQUnit 1: Ethical Frameworks for AI

Which action is an example of following an ethical framework?

Answer

(b) Testing an AI system for bias before it is deployed.

  • Checking for bias protects fairness. Using personal data without permission and hiding how decisions are made break ethical principles.
Q4 (iv)1 markMCQUnit 6: Natural Language Processing

How many tokens are in the sentence “AI, ML and NLP are changing the world.”?

Answer

(c) 10.

  • Tokenisation counts each word and each punctuation mark: AI | , | ML | and | NLP | are | changing | the | world | . — eight words and two punctuation marks make 10 tokens.
Q4 (v)1 markAssertion–reasonUnit 2: Advanced Concepts of Modeling in AI

The assertion says supervised learning does not suit clustering tasks. The reason says clustering algorithms need labelled data to group similar points.

Answer

(c) The assertion is true but the reason is false.

  • Clustering is unsupervised: it groups unlabelled data by similarity. That is exactly why supervised learning, which needs labels, is the wrong tool, and why the reason is false.
Q4 (vi)1 markNumerical MCQUnit 3: Evaluating Models

A new medical test was run 200 times. It correctly found the disease 80 times, raised a false alarm 30 times and missed the disease 40 times. What is its precision?

Answer

(a) 0.73.

  • True positives TP = 80; false positives FP = 30.
  • Precision = TP / (TP + FP) = 80 / 110 = 0.727, which rounds to 0.73.
  • The 40 missed cases are false negatives. They lower recall (80 / 120 ≈ 0.67), not precision.

Q5: subject objective questions, set 4 (answer any 5)

Q5 (i)1 markMCQUnit 6: Natural Language Processing

What is keyword extraction meant to do?

Answer

(b) Pick out the words or phrases that best represent what a text is about.

  • It does not translate, remove stopwords or write a summary; it finds the terms that capture the main topics.
Q5 (ii)1 markMCQ with imageUnit 5: Computer Vision

A picture shows an elephant, then a camera, then a laptop. Which order of stages does it show?

Answer

(c) Real object → sensing device → interpreting device.

  • The camera senses the real scene, and the computer then interprets the image it captured.
Q5 (iii)1 markMCQ (find the false statement)Unit 3: Evaluating Models

Which statement about the train-test split is not true?

Answer

(b) The claim that it can be used with any unsupervised learning algorithm.

  • The split scores a model’s predictions against the known answers in the test set, which is how supervised models are judged.
  • The other three statements are true: it needs a reasonably large dataset, it divides the data into a training part and a testing part, and it estimates how the model will do on new data.
Q5 (iv)1 markNumerical MCQ (perceptron)Unit 2: Advanced Concepts of Modeling in AI

A perceptron helps a homeowner decide on solar panels. A diagram gives four yes/no inputs with their weights, and the bias is −0.2. Find y = w1x1 + w2x2 + w3x3 + w4x4 + b.

Answer

(b) 1.5.

  • Inputs from the diagram: enough sunlight = 1, rebates available = 1, raises the home’s value = 0, environmental benefit = 1.
  • Weights from the diagram: 0.7, 0.4, 0.3 and 0.6.
  • y = 0.7×1 + 0.4×1 + 0.3×0 + 0.6×1 + (−0.2) = 0.7 + 0.4 + 0 + 0.6 − 0.2 = 1.5.

Note: CBSE notes under this question that perceptron questions will be asked in this objective format.

Q5 (v)1 markMCQUnit 1: AI Project Cycle

What happens during the data exploration stage?

Answer

(c) The data is cleaned, visualised and understood.

  • Exploration comes before modelling: you check the data, clean it and draw graphs to see its patterns.
Q5 (vi)1 markNumerical MCQUnit 3: Evaluating Models

A price model predicted ₹19,200 for a smartphone that actually costs ₹20,000. What are the error rate and the accuracy?

Answer

(d) 0.04 and 0.96.

  • Error = |actual − predicted| / actual = |20,000 − 19,200| / 20,000 = 800 / 20,000 = 0.04.
  • Accuracy = 1 − error = 1 − 0.04 = 0.96, or 96%.

Q6–Q10: Employability Skills short answers (answer any 3)

Q62 marksShort answerCommunication Skills-II

Name the types of verbal communication.

Answer

Verbal communication uses spoken or written words. Its types are interpersonal (one person to another), small-group, public (one speaker to an audience) and written communication.

Note: The marking scheme gives half a mark for each type named.

Q72 marksShort answerSelf-Management Skills-II

Explain SMART goal setting.

Answer

SMART goals are Specific, Measurable, Achievable, Relevant and Time-bound: clear, trackable targets with a deadline, for example “finish two practice papers by Sunday”.

Q82 marksShort answerEntrepreneurial Skills-II

What qualities does an entrepreneur need?

Answer

Any four, at half a mark each: confidence, patience, hard work, creativity, readiness to take risks, sound decision-making, taking responsibility for mistakes, and not giving up when problems arise.

Q92 marksShort answerICT Skills-II

How does a Trojan horse virus work?

Answer

It poses as useful, harmless software so that the user installs it. Once it runs, it can damage files, steal information or let an attacker control the computer remotely, without the user knowing.

Q102 marksShort answerGreen Skills-II

Why is education important for sustainable development?

Answer

It shows people how their choices affect the environment, so they use resources carefully, cut waste and protect nature, meeting today’s needs without harming future generations.

Q11–Q16: subject short answers (answer any 4)

Q112 marksShort answerUnit 1: Ethical Frameworks for AI

What is bioethics? Name any two of its principles.

Answer

Bioethics is the ethical framework for health care, medicine and the life sciences; for AI it guides tools such as diagnosis systems. Principles (any two): respect for autonomy, non-maleficence (do no harm), beneficence (the greatest benefit for all) and justice.

Note: The marking scheme gives 1 mark for the explanation and half a mark for each principle.

Q122 marksShort answer with exampleUnit 2: Advanced Concepts of Modeling in AI

What is an association rule in machine learning? Use an example to show how it finds links between products.

Answer

An association rule is an unsupervised method that finds items which often occur together, in the form “if A, then B”. Example: if most shoppers who buy bread also buy butter, a shop can shelve them together or suggest butter at the checkout.

Q132 marksShort answer with exampleUnit 2: Advanced Concepts of Modeling in AI

What is anomaly detection? Give an example.

Answer

Anomaly detection finds data points that do not fit the usual pattern. Example: a fitness band flags a sudden, unusual jump in heart rate, or a bank flags a card payment far outside the owner’s normal spending.

Q142 marksShort answerUnit 3: Evaluating Models

Why do we need to evaluate a model, and why should it be tested on data it has not seen?

Answer

Evaluation shows how well a model works, where it is strong or weak, and whether it suits the task. It must use unseen test data because a model tested on its own training data may just have memorised it (overfitting) and will look better than it really is.

Q152 marksShort answer (difference)Unit 5: Computer Vision

Differentiate between object detection and instance segmentation.

Answer

Object detection finds each object in an image and marks it with a labelled bounding box. Instance segmentation goes further: it labels the exact pixels of every separate object, so two people standing side by side get two separate outlines rather than one box each.

Note: CBSE’s marking scheme prints a comparison of computer vision with image processing against this question. Answer the question as printed; the two definitions above follow CBSE’s Class 10 handbook.

Q162 marksCase-based short answerUnit 6: Natural Language Processing

A newspaper uses AI to sort articles into sports, politics, health and technology. Name this NLP application and explain how it helps.

Answer

Text classification. The model reads each article and assigns it a category, which saves editors time, keeps sections organised and helps readers find the stories they want.

Note: One mark for naming the application, one for the explanation.

Q17–Q21: long answers (answer any 3)

Q174 marksCase studyUnit 1: Ethical Frameworks for AI

A school wants AI cameras that read students’ faces and body language and alert teachers to signs of bullying or stress. Some families welcome it; others object to constant watching. (a) Which ethical concerns need discussing first? (b) How can the system protect privacy, and should it act without a human checking?

Answer

(a) Privacy and dignity, consent, fairness, and accountability with transparency. (b) Collect as little as possible, keep it safe, and let the AI only raise alerts that a teacher checks.

  • (a) Privacy and dignity: students would be filmed all day. Consent: students and parents should be told clearly and asked. Fairness: reading expressions can misjudge some students, so the system must be tested for bias. Accountability and transparency: who sees the alerts, what is stored, and who answers for mistakes.
  • (b) Collect only what safety needs, store it securely for a short time, restrict who can see it and use it for nothing else. The AI should not decide anything by itself: a teacher or counsellor checks every alert and decides what happens, so no student is judged on the system’s word alone.

Note: The marking scheme awards 1 mark per point: two points for (a) and two for (b).

Q184 marksLong answerUnit 2: Advanced Concepts of Modeling in AI

You are building a smart assistant that recognises objects in photos, such as a dog or a tree, using a neural network. Explain what a neural network is and the role of its input, hidden and output layers.

Answer

A neural network is a model made of layers of connected nodes, loosely inspired by neurons in the brain; the image passes through input, hidden and output layers, and training adjusts the weights until its answers are right.

  • What it is (1 mark): layers of connected nodes, each passing its result to the next layer, loosely modelled on how brain neurons work.
  • Input layer (½ mark): takes in the image as numbers, one value per pixel, and passes them on without processing them.
  • Hidden layers (2 marks): each node combines its inputs using weights and a bias and passes the result through an activation function. Earlier layers respond to simple features such as edges and colours, later ones to shapes and parts such as ears or leaves. In training, the network compares its answer with the correct label and adjusts the weights to reduce the error, again and again, which is how it learns.
  • Output layer (½ mark): gives the final answer, for example how likely the photo is to show a dog or a tree.
Q194 marksApplied long answerUnit 2: Advanced Concepts of Modeling in AI

A school wants a chatbot on its website for everyday questions such as school timings and exam dates. (a) Should it be rule-based or learning-based? (b) Explain your choice. (c) What are its limits here? (d) How would you improve it once students start asking new kinds of questions?

Answer

(a) Rule-based. (b) Fixed rules map known questions to stored answers. (c) It cannot learn or handle questions outside its rules. (d) Add a learning-based model trained on real questions.

  • (a) A rule-based approach.
  • (b) The developer writes the rules: when a question matches a known pattern, such as “timings”, the bot returns the stored answer. That suits a short list of fixed, factual questions.
  • (c) Its learning is static. It cannot answer anything outside its rules or cope with unusual phrasing, and it does not learn from its mistakes, so every new question needs a person to add a rule.
  • (d) Switch to, or add, a learning-based model trained on real student questions so it can handle new ones, and log the questions it fails on so staff can review them.
Q204 marksNumericalUnit 3: Evaluating Models

A sentiment analyser labelled 500 responses as positive (1) or negative (0), with TP = 150, FP = 100, TN = 200 and FN = 50. Calculate accuracy, precision, recall and F1 score.

Answer

Accuracy 0.70 (70%), precision 0.60, recall 0.75, F1 score ≈ 0.67.

  • Accuracy = (TP + TN) / total = (150 + 200) / 500 = 350 / 500 = 0.70.
  • Precision = TP / (TP + FP) = 150 / (150 + 100) = 150 / 250 = 0.60.
  • Recall = TP / (TP + FN) = 150 / (150 + 50) = 150 / 200 = 0.75.
  • F1 = 2 × precision × recall / (precision + recall) = 2 × 0.60 × 0.75 / 1.35 = 0.90 / 1.35 ≈ 0.667.

Note: Half a mark for each formula and half for each answer, so write every formula.

Q214 marksApplied long answer (TF-IDF)Unit 6: Natural Language Processing

The paper gives three very short documents: one about a cat sitting on a mat, one about a dog that is barking, and one saying both are good pets. (a) What is term frequency, and what is it here? (b) Build the document frequency table. (c) Build the inverse document frequency table. (d) Give the TF-IDF formula for a word W.

Answer

See the tables below. In CBSE’s convention IDF = total documents ÷ document frequency, and TFIDF(W) = TF(W) × log(IDF(W)) with log to base 10.

  • (a) Term frequency is the number of times a word occurs in one document. Document 1 uses “the” twice and each of its other words once; documents 2 and 3 use each of their words once.
  • (b) Document frequency counts how many documents contain a word, however often: “the” appears in 2 documents; each of the other 11 words appears in 1.
  • (c) IDF = 3 ÷ DF: 3/2 for “the” and 3/1 for every other word.
  • (d) TFIDF(W) = TF(W) × log(IDF(W)). Example: “the” in document 1 scores 2 × log 1.5 ≈ 0.35, while “cat” scores 1 × log 3 ≈ 0.48, so the rarer word carries more weight.

Note: One mark for each table and one for the formula.

Q21 (a)–(c): term frequency, document frequency and IDF for the three documents
WordTF in doc 1TF in doc 2TF in doc 3DFIDF (3 ÷ DF)
the21023/2
cat10013/1
sat10013/1
on10013/1
mat10013/1
dog01013/1
is01013/1
barking01013/1
both00113/1
are00113/1
good00113/1
pets00113/1

From: Artificial Intelligence Facilitator Handbook, Class X (Subject Code 417), Unit 6, TFIDF section

Where does CBSE’s marking scheme not match the paper?

How should a student practise with this paper?

  1. Sit the paper first, timed at two hours, without looking at any answers.
  2. Mark it with CBSE’s marking scheme, then read our explanation for every question you lost marks on.
  3. Re-do every numerical from scratch a day later: the confusion-matrix metrics, single-prediction error and accuracy, the perceptron sum and TF-IDF.
  4. For each short answer, check the length: 20–30 words for 2 marks, 50–80 for 4 marks. Examiners mark points, not length.
  5. Move on to our Class 10 question bank for fresh questions in the same formats.

Frequently asked questions

Where is the 2026-27 Class 10 AI (417) sample paper?
CBSE hosts the 2026-27 sample paper and marking scheme for 417 on its skill education portal, skilleducation.cbseit.in, and this page links both PDFs. On 3 November 2026 the skill-subject sample paper page on cbseacademic.nic.in still linked the 2025-26 files.
How many questions are compulsory in the Class 10 AI paper?
Students answer 15 of the 21 questions: all five in Section A, choosing among the parts inside each, and 10 of the 16 in Section B.
What is the answer to Q2 (iv) in the 2026-27 Class 10 AI sample paper?
As printed, option (b): a high F1 score shows a good balance between precision and recall. The marking scheme’s line for this question names an option that is not in the paper.
Which numericals are in the 2026-27 Class 10 AI sample paper?
Precision from a medical-test case (Q4 vi), a perceptron’s weighted sum (Q5 iv), error and accuracy for one price prediction (Q5 vi), all four classification metrics from a confusion matrix (Q20) and term, document and inverse document frequency (Q21).

Sources and versions

We link to the documents as their publishers host them. We read each file in full on the date shown; open the publisher's PDF to check anything on this page.

Related guides

Parents: start your child free

LittleAIMaster is an AI course app for Grades 6–12 with short, interactive lessons on how AI works. It is free to start, and you decide whether to upgrade.

Teachers: free AI worksheets for your class

Printable AI worksheets for Grades 6–12 that you can use in a lesson today.

LittleAIMaster is an independent learning app. It is not affiliated with or endorsed by CBSE, NCERT or any government education board. We link to the documents CBSE and NCERT publish rather than copying them, and quote them only briefly, with the source named.