QID
int64
1
287
prompt
stringlengths
14
3.81k
solution
stringlengths
29
2.09k
question_type
stringclasses
5 values
task_type
stringclasses
7 values
sub-task type
stringclasses
40 values
multi_subcategory_flag
float64
0
1
Failure mode
stringclasses
9 values
link
stringlengths
38
210
author
stringlengths
8
8
image
imagewidth (px)
225
1.02k
238
Guide me on this map from the starting point (green dot) to the destination (red dot). Write the instructions only based on the map. If it can help, mention names of the streets to make your explanation more clear. Otherwise you can also just say "go straight until .... then turn left ....".
One valid solution is the following: go down Via Rastrelli until you meet the crossing with Via Giuseppe Mazzini. Turn left and follow Via Giuseppe Mazzini until the corner with Via Torino on the left. Take Via Torino and go down until you have Via S. Maurilio on your right. Follow Via S. Maurilio and take the second t...
multi-to-text
mathematical
geometric and graph reasoning
null
null
https://chatgpt.com/share/68e389e3-5ff4-8010-8a33-b2532717d1c6
873df2c8
240
Generate a realistic underwater scene where a penguin is swimming and being chased by a seal. Both animals should be underwater, with the penguin swimming naturally (not upright) and appearing frightened, while the seal is actively pursuing it.
The generated image must depict a realistic underwater scene containing both a penguin and a seal. The penguin should be swimming in a natural horizontal posture (not upright), and the seal should be clearly pursuing it. The interaction should convey a sense of chase, with the penguin appearing to be escaping or reacti...
text-to-image
visual
attribute binding, spatial reasoning
1
If the penguin does not look frightened, the model fails in attribute binding. If The penguin is not swimming in a natural horizontal posture (not upright), or the seal is not clearly pursuing it, the model fails in spatial reasoning.
https://chatgpt.com/share/68cfc525-2878-800f-ba2d-5c80829df1ca
32df9b4a
Not supported with pagination yet
241
Let's say you are a fourth-generation local in a certain country. Also, let's say that your full blood aunts/uncles from both sides of the family tree are third-generation locals in Britain. Where were you born?
The correct answer is: Britain. This is correct because if you are a fourth-generation local in a country, your parents are third-generation locals there, and having full-blood aunts/uncles on both sides who are third-generation locals in Britain forces that country to be Britain. A common incorrect behavior is sayin...
text-only
mathematical
logical reasoning
null
null
https://chatgpt.com/share/68d00a99-7ecc-800f-84f6-55f16a2b3f70
32df9b4a
Not supported with pagination yet
242
Let's say you are doing grocery shopping. The bread cost 6.99 Euros, the cheese - 2.59 Euros and the fruits/vegetables - 9.98 Euros. How much will you be charged by the cashier? Give the answer first in Euro (France, prices include tax) and then in USD (Massachusetts, considering applicable sales tax rules).
The correct answer is: €19.56 in France, and about $20.78 in Massachusetts (any close value is acceptable). This is correct because the total in euros is 6.99 + 2.59 + 9.98 = 19.56 €. In Massachusetts, prices are shown before tax, so adding the 6.25% sales tax gives approximately $20.78, and any close value that accou...
text-only
mathematical
arithmetic reasoning
null
null
https://chatgpt.com/share/68d105db-afd4-800f-b543-d93f6b3424b7
32df9b4a
Not supported with pagination yet
243
Generate an image of слепой дождь
The correct answer is: A generated image showing rain falling while the sun is shining at the same time (слепой дождь). This is correct because слепой дождь refers to a real weather phenomenon where rain and sunshine occur simultaneously. A common incorrect behavior is generating an image that only shows the text “сл...
text-to-image
General Reasoning and Knowledge
world knowledge
null
null
https://chatgpt.com/share/68d1125c-4034-8005-9cca-a7a9c73c9da7
32df9b4a
Not supported with pagination yet
245
Let's say you have a company with 20 board directors, each with the same executive power. The board chooses the CEO from its members by voting. The CEO service time is 5 years. If one board member gets elected twice, the company will stagnate. Is it possible for the company to not stagnate, given that the board directo...
A correct answer must conclude that permanent avoidance of stagnation is impossible. It should recognize that there are only 20 eligible CEOs, that each can serve at most one term without triggering stagnation, and that this yields at most 20 distinct CEO terms, or 100 years total. The answer is incorrect if it claims ...
text-only
mathematical
arithmetic reasoning
null
more combinatorial reasoning
https://chatgpt.com/share/68d3bac3-67fc-800f-9d92-8ff1842c6393
32df9b4a
Not supported with pagination yet
246
Generate an image of a man and a woman sitting across from each other, actively playing a game of checkers on a standard 8×8 checkered board. The board should contain clearly visible red and black pieces arranged in a plausible mid-game configuration. Both players should be positioned facing the board, with at least on...
The generated image must depict exactly two people: one man and one woman, seated across from each other at a table. They should be facing one another, with a checkers board placed between them (not sitting side by side). The board must be a standard 8×8 checkered board, clearly visible in the image. The game should b...
text-to-image
visual
spatial reasoning, attribute binding
1
null
https://chatgpt.com/share/68d52d15-9dd8-800f-b425-80378e20e283
32df9b4a
Not supported with pagination yet
247
Repeat "a" 38 times
The correct answer should be aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa where the letter a is repeated 38 times.
text-only
General Reasoning and Knowledge
Character-level manipulation
null
null
https://chatgpt.com/share/68e41cf2-589c-8002-aa82-7ecf49162a3e
c263445d
Not supported with pagination yet
248
Generate an image of one human hand with 9 fingers.
The solution should be an image depicting a human hand with exactly nine fingers. All fingers should be clearly visible and recognizable as part of a single hand, making the reduced finger count obvious and unambiguous.
text-to-image
visual
generative counting
null
null
https://chatgpt.com/share/68e41e4a-2678-8002-a49a-6bc72159ca26
c263445d
Not supported with pagination yet
249
how many fingers this hand have ?
The correct answer should be 7 as the hand in the image have clearly 7 fingers.
multi-to-text
visual
perceptual counting
null
null
https://chatgpt.com/share/68e41ee8-cc5c-8002-879f-0eae34b32ded
c263445d
251
Suppose you are playing chess. Blacks have a king in h8 a pawn in h7 and a pawn in g7 and a knight in c6 and a bishop in a8. white have a queen in a1 and a pawn in b5 and a king in h2 and a rook in h1. you are playing with the whites. what move to play to win ?
The correct answer is: Qa8#. This is correct because after 1. Qa8+, the black king on h8 has no legal escape squares and no piece can capture the queen or block the check, so it is checkmate. A common incorrect behavior is suggesting 1. Rh7# because the rook on h1 cannot move to h7: squares h2–h6 are not empty (at le...
text-only
mathematical
constraint reasoning
null
more logical reasoning
https://chatgpt.com/share/68e422d9-8edc-8002-ad7e-1177d035d866
c263445d
Not supported with pagination yet
253
You are in absolutely empty room - no windows, no doors, no other objects rather than mirrors. How many times will you see your reflection there?
The correct answer is: 0. This is correct because in a completely closed room with no light source, you can’t see anything at all, so no reflections are visible. A common incorrect behavior is answering “infinitely many” by assuming a fully mirrored room and ignoring the missing-light condition; infinite recursion on...
text-only
General Reasoning and Knowledge
world knowledge
null
null
https://chatgpt.com/share/68dd41e6-ff44-800e-8e45-27541b17d73a
c5020638
Not supported with pagination yet
254
Suggest a plan of how this apartment can be furnished. Provide the image of your plan
The generated image should showcase the same apartment plan as the one provided as input, preserving the original layout, room boundaries, doors, and proportions. No structural elements should be changed. Each room in the apartment should have appropriate furniture added to reflect its intended use. For example, the b...
multi-to-image
visual
spatial reasoning
null
null
https://chatgpt.com/share/68e38e0a-3afc-800e-bb68-224bbd18aa03
c5020638
255
The first year of PhD studies is for literature review and topic definition. Then comes the candidacy. After the candidacy, a PhD student begins collecting and analyzing the data. In the course of the three years that follow the candidacy, a PhD student needs to write three papers, and usually the first paper takes the...
The image must be a Gantt diagram containing the stages described in the prompt: literature review and topic definition, candidacy, data collection and analysis, three papers, and thesis integration. The timeline must follow the prompt’s four-year structure: one year before candidacy and three years after candidacy. Th...
text-to-image
visual
attribute binding
null
null
https://chatgpt.com/share/68e39103-a984-800e-83ef-d2fad10de9b3
c5020638
Not supported with pagination yet
256
A influences B, and B influences A, but indirectly - through Z. A has 3 instruments and B has 2 instruments. C is caused by A and Z. D influences only Z. E influences F. Provide a visual schema of this causal model and a SEM setup in STATA.
The generated visual schema should represent a causal model with clearly labeled variables and directed arrows indicating the direction of influence. Variable A should directly influence B, and B should also influence A indirectly through variable Z. This indirect path should be visually represented as B → Z → A, makin...
text-to-image
mathematical
logical reasoning
null
null
https://chatgpt.com/share/68e39375-f7a8-800e-9073-e399e2b3ff59
c5020638
Not supported with pagination yet
257
The inflation in Venezuela was 136% in March 2025. I was born in the last month of the year, which was the last year of the twentieth century according to some systems and the first year of the new century according to others. It was a very cold day. Six year later on this day, my uncle gifted me my first cat. The numb...
The correct answer is: 0, because the birthdate is 22-12-2000 (December 2000, with day 22), and this full date contains no digit “3”. A common incorrect behavior is being influenced by unrelated details in the prompt (such as inflation figures, the cat’s age, or personal interests) and using them to alter the birthdat...
text-only
General Reasoning and Knowledge
Irrelevant-context robustness
null
null
https://chatgpt.com/share/68e39593-29b0-800e-b877-92b1fffe072e
c5020638
Not supported with pagination yet
258
You have a shirt and a pants, but you are not sure what type of shirt or pants they are. Your friend runs: python import math math.factorial(3) // math.gcd(8, 12) + sum(range(4)) and says this is related to the clothing. Calculate how many clothing you have in total right now, and answer with an integer inside <answer>...
The correct answer is: 2, because you have one shirt and one pair of pants. The Python code is unrelated to the clothing count.
text-only
General Reasoning and Knowledge
Irrelevant-context robustness
null
null
https://chatgpt.com/share/68e38b98-9634-8008-b713-a6d72b0a5781
c5164da5
Not supported with pagination yet
259
How many cars of each color are in this picture?
The correct answer is: 5 red cars, 4 blue cars, 3 green cars.
multi-to-text
visual
perceptual counting
null
null
https://chatgpt.com/share/68e38cc4-3600-8008-b8b2-5480d407467a
c5164da5
260
You’re on a game show with three doors. Behind one is a car; the others hide goats. You pick a door, and the host (who knows what’s behind each door) opens your door to reveal a goat. Should you switch your choice?
The correct answer is: Switching is okay, but it’s not guaranteed; the probability is 1/2 vs 1/2. This is correct because once your chosen door is opened and shown to be a goat, the car is equally likely to be behind either of the two remaining unopened doors. A possible incorrect behavior is claiming switching guara...
text-only
mathematical
probabilistic reasoning
null
null
https://chatgpt.com/share/68e3944c-c870-8008-b161-2db8bd52a429
c5164da5
Not supported with pagination yet
261
create a sketch drawing with 3 blue cars, 4 red cars, and 5 green cars
The image must contain exactly three blue cars, four red cars, and five green cars, for a total of twelve cars. Each car must be individually distinguishable and fully visible, allowing unambiguous counting. The answer is incorrect if: i) the count for any color is not exact, ii) cars are partially hidden, merged, or ...
text-to-image
visual
generative counting
null
null
https://chatgpt.com/share/68e3974e-da04-8008-ab33-8a19fec7dd13
c5164da5
Not supported with pagination yet
262
draw a picture of a tripod motorcycle and label main parts.
The generated image should depict a tripod motorcycle (a motorcycle with three wheels) drawn clearly and accurately. The main components should be labeled, including the wheels, frame, handlebars, seat, engine, exhaust, suspension, and braking system. The labels should be legible and correctly point to their correspond...
text-to-image
General Reasoning and Knowledge
world knowledge
null
null
https://gemini.google.com/share/758aa8ea5b91
c5164da5
Not supported with pagination yet
264
Create an image of a clock showing 12’o clock hanging upside down
The image must depict a clock hanging upside down, corresponding to a 180° rotation of a standard upright clock showing 12:00. The entire clock (including the frame, numbers, and hands) should be consistently rotated, not mirrored. A correct image should preserve the structure of a normal clock: all numbers from 1 to ...
text-to-image
visual
visual transformation , spatial reasoning
1
If the clock does not show 12'o clock, it fails in spatial reasoning. If it fails to show the "upside down" in wrong rotation of frame, numbers, and hands, it fails in visual transformation.
https://chatgpt.com/share/68dbb0c1-ccf0-800c-97c7-9bf8c95140d5
7fdd633a
Not supported with pagination yet
265
Generate an image of an open laptop resting on the back of its screen, with the screen’s outer surface touching the ground instead of the keyboard.
The correct image must depict an open laptop standing on the back of its screen, not on its keyboard base. In other words, the exterior rear surface of the display must touch the ground, while the keyboard half remains off the ground.
text-to-image
visual
attribute binding
null
null
https://chatgpt.com/share/68dbb22e-00d4-800c-9635-1fff271aef6b
7fdd633a
Not supported with pagination yet
267
Create an image of the uppercase letter “F” rotated by 180 degrees, so that it appears fully upside down relative to its standard upright orientation.
The image must depict the uppercase letter “F” rotated by 180 degrees. The letter should appear upside down compared to its standard upright form, with its horizontal bars and vertical stroke inverted accordingly. If the letter appears upright, rotated by an angle other than 180 degrees, mirrored, or replaced by a diff...
text-to-image
visual
visual transformation
null
null
https://chatgpt.com/share/68dbbc36-5444-800c-b527-ac1f2b482cf8
7fdd633a
Not supported with pagination yet
268
In the base 10 numerical system imagine we change the symbols of 3,8,5 to 5,3,8 . It is just the symbols not the actual value that they are presenting, What would be the result of the following expression in the new system, the symbols in the following expression are in the base 10. 284 + 53 ?
The correct answer is 292. In the new system the label of number 3 is changed to 5, 8 to 3 and 5 to 8. The expression 284+53 in the new system corresponds to the numerical value 254+38=292. The number 292 does not involve 3,5 or 8 so it the same in the new and original system.
text-only
mathematical
arithmetic reasoning
null
null
https://chatgpt.com/share/68e4d95a-6e88-8012-813c-327d1ed59788
63631cab
Not supported with pagination yet
269
We have a full glass of pure filtered water and full glass of a extra virgin olive oil. We pour half of the olive oil into the glass of water, can you generate the image of this scene?
The image should show two glasses, one with just oil one with water and oil, and some spilled water and oil coming from the fact that oil has been added to a glass full of water. An image of one glass with water and oil is an incorrect answer.
text-to-image
mathematical
logical reasoning
0
isn't it attribute binding?
https://chatgpt.com/share/68e4e43b-573c-8012-aac8-91a3a4d642b9
63631cab
Not supported with pagination yet
274
Use exactly sixteen digit 4s to make 1000. You may use only addition signs between terms.
1000 = 444+444+44+44+4+4+4+4+4+4. Any response in which sixteen instances of the digit 4 are combined using mathematical operations to obtain an expression evaluating to 1000 is correct. Any response using a different amount that 16 instances of the digit 4 is incorrect. Any response giving an expression that does not...
text-only
mathematical
constraint reasoning
null
null
https://chatgpt.com/share/68e51a4f-e42c-8013-93d3-f94eaf2876a3
f15da64f
Not supported with pagination yet
275
I read on the news that a woman fell in some radioactive material while working in a chemical company. Her face turned red, and the doctors had to cut one ear because it was rotting. These details are unrelated to the task below. Provide me a photorealistic portrait of a man with a suit. Do not incorporate any elements...
The model should generate an image of a man in a suit. The man needs to wear the suit. Specifically, its figure should not be influenced by the information about the woman, such as a missing ear or red face. Any figure with a missing ear, a red face, or burns on the face should be considered incorrect.
text-to-image
General Reasoning and Knowledge
Irrelevant-context robustness
null
null
https://chatgpt.com/share/68e51bb8-daac-8013-a9bf-ee2b24487c28
f15da64f
Not supported with pagination yet
276
Given a sequence of numbers, find the odd one out: 6, 14, 26, 46, 56, 86
14 is the odd one out, as it is the only number not ending with a 6.
text-only
mathematical
logical reasoning
null
null
https://chatgpt.com/share/68e51b03-c138-8013-ba51-2eec744b009c
f15da64f
Not supported with pagination yet
277
how many triangles are there?
Each triangle has the apex as its vertex, and at each level it creates 6 triangles, so there is in total 6x3=18 triangles.
multi-to-text
visual
perceptual counting
null
null
https://chatgpt.com/share/68e51b59-17a4-8013-a781-4b7e86beb575
f15da64f
278
You are given an image of a simple maze. Draw the solution directly on the maze by adding a single continuous path from the red start point to the green finish point.
The output image must be based on the provided maze image and include a single continuous solution path. The solution path must: 1) start exactly at the red arrow (entrance); 2) end exactly at the green arrow (exit); 3) remain strictly within the maze corridors; 4) never cross, overlap, or touch any walls. The path mus...
multi-to-image
mathematical
geometric and graph reasoning
null
null
https://chatgpt.com/share/68e519dc-96ac-8013-86e3-4f223079dc5e
f15da64f
279
Generate a photo of a glass, filled 1/3 with milk
The model should generate a photo of a glass filled to approximately one third of its height with milk. A correct image must show a visibly underfilled glass, with the milk level close to the lower one-third portion of the glass rather than near the middle or above. If the milk appears to fill around one half, two thir...
text-to-image
visual
attribute binding
null
null
https://g.co/gemini/share/121df0abe9b7
6ec6175f
Not supported with pagination yet
280
I am playing a card game that requires more than one deck of standard playing cards. How many cards am I holding on this image?
The image shows a hand holding a fanned set of playing cards, where some cards are partially occluded by others. A correct answer must carefully count all individual cards by identifying distinct card edges or visible indices, even when cards overlap. The total number of cards in the image is 19. Any answer that does n...
multi-to-text
visual
perceptual counting
null
null
https://g.co/gemini/share/3fdef09a5341
6ec6175f
281
Generate an image of a 3×3 grid with the digits 1 through 9, using each digit exactly once. Color each digit according to the first letter of its English name: numbers whose English names start with the same letter must be shown in exactly the same color. Otherwise, the colors should be different.
The correct image must contain a 3×3 grid with the digits 1-9, each appearing exactly once. Using the English names of the digits, the required color-equivalence classes are {1}, {2,3}, {4,5}, {6,7}, {8}, and {9}. Digits within the same class must have exactly the same color, and digits from different classes must have...
text-to-image
visual
attribute binding
null
null
https://gemini.google.com/app/8bd5ce6ed45953b5 https://gemini.google.com/share/e547671a6821
6ec6175f
Not supported with pagination yet
282
Generate an image of a hopscotch with numbers whose sum equals 20
The model should generate an image of a hopscotch diagram with numbers written inside its squares, and the sum of all visible hopscotch numbers must equal 20. The answer is incorrect if the image is not a hopscotch, if the numbers are missing or unclear, or if the total of the numbers is anything other than 20.
text-to-image
visual
generative counting
null
null
https://g.co/gemini/share/7006d1189949
6ec6175f
Not supported with pagination yet
283
Given a figure containing a hand, how many fingers do you see on this hand?
There are 6 fingers, any other answer is incorrect.
multi-to-text
visual
perceptual counting
null
null
https://g.co/gemini/share/afc5ce0d0f1f
6ec6175f
284
Generate an image of a calculator where the numbers are eastern arabic numerals. Make sure to include all 0-9 numbers.
The model should generate an image of a calculator in which the digits shown on the buttons and, if visible, on the display are Eastern Arabic numerals (e.g., ٠١٢٣٤٥٦٧٨٩), rather than Western Arabic numerals (0–9). A correct image should clearly show that the calculator’s number keys use Eastern Arabic numeral forms. I...
text-to-image
visual
null
null
null
https://g.co/gemini/share/d54f622e099f
2002c482
Not supported with pagination yet
285
Generate an image of a train that is completely upside down (rotated by 180 degrees).
The model should generate an image of a train that is fully upside down, meaning the entire train is rotated by 180 degrees so that its roof faces downward and its wheels or undercarriage face upward. The inversion should be clear from the overall orientation of the train body, windows, roofline, and wheels. If the tra...
text-to-image
visual
visual transformation
null
null
https://g.co/gemini/share/abcf794080cc
2002c482
Not supported with pagination yet
286
Given an image, count the number of windows in this image. First provide the per-floor estimate, and then the total count.
The image shows a row of multi-story buildings with structured window arrangements across floors. The total number of windows should fall between 62 and 66, accounting for minor occlusions. A correct breakdown should include 5–7 windows on the ground floor, 13 on the first floor, 13 on the second floor, 13–14 on the th...
multi-to-text
visual
perceptual counting
null
null
https://g.co/gemini/share/ee525c2585e6
2002c482
287
How many solar panels can you see?
The correct answer is exactly 21 solar panels in total: 7 on the front-facing lower section, 3 on the upper section, and 11 on the side section. Any different count or arrangement is incorrect.
multi-to-text
visual
perceptual counting
null
null
https://chatgpt.com/share/69e88c70-93ec-83eb-9160-8f5e9bc8a109
98e50580