Datasets:
judgment_id int64 1 10.8k | snapshot_id stringclasses 1
value | prediction_id int64 1 2.86k | post_id stringclasses 519
values | model_id stringclasses 6
values | judge_model stringclasses 3
values | verdict stringclasses 2
values | reasoning stringlengths 49 2.24k | judge_prompt_id stringclasses 2
values | judged_at stringlengths 32 32 | is_latest bool 2
classes |
|---|---|---|---|---|---|---|---|---|---|---|
646 | ad24bf870ce6285d | 360 | 16uuwk3 | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke mechanism: John Cena's WWE catchphrase 'You can't see me' being taken literally as if he's invisible. The model names John Cena specifically, identifies the catchphrase, and explains why it's funny - someone pretending they can't see Cena in the photos, playing on the intern... | 4b9c000df6b40d4d | 2026-05-27T02:54:22.322425+00:00 | false |
4,028 | ad24bf870ce6285d | 360 | 16uuwk3 | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke: John Cena's catchphrase 'You can't see me' is taken literally as if he's invisible, so in photos with him, people pretend they can't see him. The model names John Cena specifically, references the catchphrase, and explains the meme mechanism correctly. The model adds the de... | 52b3fd02a1b032ad | 2026-06-20T00:46:20.019749+00:00 | true |
264 | ad24bf870ce6285d | 360 | 16uuwk3 | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the specific reference to John Cena and his catchphrase/meme "You can't see me," and explains the core joke that he is treated as invisible in photos. This matches the ground truth's mechanism: only the poster is visible while Cena is 'not seen.' The explanation also captures why it's fun... | 4b9c000df6b40d4d | 2026-05-27T02:52:07.231022+00:00 | false |
4,018 | ad24bf870ce6285d | 360 | 16uuwk3 | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the specific John Cena meme and catchphrase "You can't see me," and explains the core joke that Cena is treated as if he were invisible in photos. This matches the ground truth's setup: in a selfie/photo, only the other person is visible because Cena is 'not seen.' The model adds some ext... | 52b3fd02a1b032ad | 2026-06-20T00:46:18.040817+00:00 | true |
2,649 | ad24bf870ce6285d | 360 | 16uuwk3 | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies John Cena, his catchphrase 'You can't see me', and the meme that he is literally invisible. It also correctly explains the joke mechanism in the context of the photos, matching the ground truth. | 52b3fd02a1b032ad | 2026-06-20T00:31:38.221738+00:00 | true |
767 | ad24bf870ce6285d | 419 | 18lkyqa | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies all the key elements of the meme: the 'broccoli top' haircut associated with Gen Z boys, Monday as 'International Chest Day,' the specific clothing items (pajama/plaid pants, Crocs), the bench press/gym equipment being hogged, and the group dynamic of multiple people crowding around one p... | 4b9c000df6b40d4d | 2026-05-27T02:55:13.612259+00:00 | false |
4,272 | ad24bf870ce6285d | 419 | 18lkyqa | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model's explanation covers all the key elements of the ground truth: the Gen Z 'broccoli top' haircut stereotype, the gym setting on Monday 'International Chest Day,' the Crocs, pajama/plaid pants, and multiple guys crowding around one bench press. The model also correctly identifies the specific joke about only on... | 52b3fd02a1b032ad | 2026-06-20T00:47:28.760149+00:00 | true |
385 | ad24bf870ce6285d | 419 | 18lkyqa | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke: it is about Gen Z/high school boys with broccoli-top haircuts going to the gym on Monday, i.e. 'International Chest Day,' wearing Crocs and pajama/lounge pants, and crowding around the bench press. It also captures the key humor that these groups hog equipment and are stere... | 4b9c000df6b40d4d | 2026-05-27T02:52:28.095038+00:00 | false |
4,260 | ad24bf870ce6285d | 419 | 18lkyqa | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke: Gen Z/high school boys with broccoli hair going to the gym on Monday, i.e. 'International Chest Day,' wearing Crocs/pajama pants and crowding around the bench press. It also captures the idea that they gather in groups and hog equipment. However, the ground truth specifical... | 52b3fd02a1b032ad | 2026-06-20T00:47:26.025709+00:00 | true |
2,826 | ad24bf870ce6285d | 419 | 18lkyqa | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies all the key elements from the ground truth: the 'broccoli top' haircut associated with Gen Z/high school boys, Monday as 'International Chest Day,' the stereotypical outfit (Crocs, pajama/lounge pants), and the joke about groups crowding around the bench press with only one person working... | 52b3fd02a1b032ad | 2026-06-20T00:34:02.082963+00:00 | true |
747 | ad24bf870ce6285d | 410 | 1966nvq | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke mechanism: the lobsters are in a tank on the Titanic, and when the ship sinks, the lobsters ironically survive/escape their fate as dinner. The model correctly identifies the Titanic reference, understands the ironic twist, and explains why it's funny (the lobsters escape be... | 4b9c000df6b40d4d | 2026-05-27T02:55:05.185945+00:00 | false |
4,222 | ad24bf870ce6285d | 410 | 1966nvq | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke: the lobsters are in a tank on the Titanic, and the ship is about to sink due to the iceberg, which means the lobsters will 'escape' being eaten because of the disaster. The model mentions the Titanic specifically, the iceberg, and the ironic twist that the lobsters survive ... | 52b3fd02a1b032ad | 2026-06-20T00:47:14.876036+00:00 | true |
363 | ad24bf870ce6285d | 410 | 1966nvq | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke mechanism and the specific reference to the Titanic. It explains that the lobsters are spared from being eaten because the ship is about to hit an iceberg and sink, which matches the ground truth’s irony: the disaster prevents the humans from killing/eating them, allowing th... | 4b9c000df6b40d4d | 2026-05-27T02:52:24.373427+00:00 | false |
4,216 | ad24bf870ce6285d | 410 | 1966nvq | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke: the lobsters are on the Titanic, and the impending iceberg/sinking prevents them from being killed and eaten, allowing them to survive and return to the ocean. This matches the ground truth’s ironic setup and outcome. It also mentions the Titanic explicitly, which is the ke... | 52b3fd02a1b032ad | 2026-06-20T00:47:12.900481+00:00 | true |
2,805 | ad24bf870ce6285d | 410 | 1966nvq | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the key elements: the lobsters in a tank awaiting being eaten, the Titanic setting, the iceberg, and the irony that the Titanic disaster spares the lobsters from being killed/eaten by humans. It matches the ground truth's core joke mechanism. | 52b3fd02a1b032ad | 2026-06-20T00:33:45.861281+00:00 | true |
741 | ad24bf870ce6285d | 407 | 1ao0u0a | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke mechanism: Church's Chicken is known for operating in rough neighborhoods, so if a Church's location is boarded up/closed, it signals the neighborhood is extremely dangerous—even worse than what Church's normally tolerates. The model captures the same hyperbole and subverted... | 4b9c000df6b40d4d | 2026-05-27T02:55:02.561709+00:00 | false |
4,212 | ad24bf870ce6285d | 407 | 1ao0u0a | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke: Church's Chicken is known for thriving in rough neighborhoods, so if a Church's location is boarded up/closed, it signals the neighborhood is so bad that even this chain couldn't survive there. The model gets the specific chain (Church's Chicken), the cultural reputation of... | 52b3fd02a1b032ad | 2026-06-20T00:47:11.677608+00:00 | true |
357 | ad24bf870ce6285d | 407 | 1ao0u0a | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke mechanism: hyperbole and subverted expectation around Church's Chicken being a durable presence in rough neighborhoods, so a boarded-up Church's implies the neighborhood is even worse. It also explains why that is funny. The model does not introduce a conflicting reference o... | 4b9c000df6b40d4d | 2026-05-27T02:52:23.378554+00:00 | false |
4,204 | ad24bf870ce6285d | 407 | 1ao0u0a | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke and the specific reference to Church's Chicken. It explains the same setup as the ground truth: Church's is stereotypically ubiquitous in rough neighborhoods, so seeing one boarded up implies the neighborhood is even worse than usual. It also captures the subverted expectati... | 52b3fd02a1b032ad | 2026-06-20T00:47:09.877646+00:00 | true |
2,792 | ad24bf870ce6285d | 407 | 1ao0u0a | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies Church's Chicken, its reputation for operating in rough/lower-income neighborhoods, and the core joke: if even a Church's has shut down, the area must be exceptionally bad. This matches the ground truth's explanation of subverted expectation and hyperbole. The model also adds appropriate ... | 52b3fd02a1b032ad | 2026-06-20T00:33:35.633156+00:00 | true |
755 | ad24bf870ce6285d | 414 | 1bazqft | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke mechanism: the Roman practice of decimation where 1 in 10 soldiers was killed, and the pun on 'only one in ten will get this' (a common meme phrase for 'understand this') versus 'one in ten will get [beaten/killed].' The model identifies the specific historical reference (de... | 4b9c000df6b40d4d | 2026-05-27T02:55:08.801268+00:00 | false |
4,241 | ad24bf870ce6285d | 414 | 1bazqft | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies: (1) the Roman practice of decimation where 1 in 10 soldiers was killed/beaten, (2) the wordplay on 'one in ten will get this' as a double meaning - the common meme phrase about only clever people understanding it, versus the literal meaning that one in ten soldiers gets beaten, (3) the c... | 52b3fd02a1b032ad | 2026-06-20T00:47:20.576226+00:00 | true |
371 | ad24bf870ce6285d | 414 | 1bazqft | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke as Roman decimation and explains that it was a punishment where one in ten soldiers were selected and punished, matching the ground truth’s emphasis on the 'ten' in 'decimation' and the one-tenth punishment. It also explains why that is funny as a pun on 'one in ten.' Althou... | 4b9c000df6b40d4d | 2026-05-27T02:52:26.063250+00:00 | false |
4,234 | ad24bf870ce6285d | 414 | 1bazqft | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core reference as Roman decimation: a punishment for mutiny where one in ten soldiers is selected and punished. It also captures the key joke mechanism from the ground truth, namely the play on the word 'ten' / 'one in ten' in relation to 'decimation.' Although the model adds extra de... | 52b3fd02a1b032ad | 2026-06-20T00:47:18.303434+00:00 | true |
2,816 | ad24bf870ce6285d | 414 | 1bazqft | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the Roman military practice of decimation, explains that it involved killing 1 in 10 soldiers as punishment for mutiny, and captures the wordplay on 'one in ten will get this' meaning both understanding the joke and literally receiving the punishment. This matches the ground truth's core ... | 52b3fd02a1b032ad | 2026-06-20T00:33:53.789368+00:00 | true |
757 | ad24bf870ce6285d | 415 | 1bcq3b4 | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies: (1) the specific person John Barnett by name, (2) the Boeing whistleblower controversy, (3) the implication that Boeing may have had him killed, (4) the dark/conspiratorial humor. The model also adds additional context about Joshua Dean and other details about the visual elements. The co... | 4b9c000df6b40d4d | 2026-05-27T02:55:09.824831+00:00 | false |
4,249 | ad24bf870ce6285d | 415 | 1bcq3b4 | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies: (1) the core joke about Boeing whistleblowers dying under suspicious circumstances, (2) specifically names John Barnett, (3) references the conspiracy theory that Boeing had whistleblowers killed, (4) the dark humor angle. The ground truth focuses specifically on John Barnett dying while... | 52b3fd02a1b032ad | 2026-06-20T00:47:23.181182+00:00 | true |
375 | ad24bf870ce6285d | 415 | 1bcq3b4 | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core dark joke: Boeing whistleblowers dying under suspicious circumstances and the implication that Boeing silenced critics. It specifically names John Barnett, which is the key person in the ground truth, and explains the humor as suspicious timing and possible foul play. However, it... | 4b9c000df6b40d4d | 2026-05-27T02:52:26.503935+00:00 | false |
4,239 | ad24bf870ce6285d | 415 | 1bcq3b4 | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke: a dark Boeing whistleblower death conspiracy, specifically referencing John Barnett and the implication that Boeing silenced him. This matches the ground truth’s explanation that the humor comes from Barnett allegedly dying by suicide while testifying against Boeing, with t... | 52b3fd02a1b032ad | 2026-06-20T00:47:19.928979+00:00 | true |
2,817 | ad24bf870ce6285d | 415 | 1bcq3b4 | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the core joke: the suspicious death of Boeing whistleblower John Barnett and the implication/conspiracy theory that Boeing had him killed to silence him. It also mentions Joshua Dean, another whistleblower who died. The model captures the dark humor and the conspiracy angle, matching the ... | 52b3fd02a1b032ad | 2026-06-20T00:33:54.744971+00:00 | true |
734 | ad24bf870ce6285d | 404 | 1bcq7up | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies all the key elements: the Mafia neighbor, paying a child $10 to start his car, the dark irony that the child was being used as a disposable bomb tester in case the car was rigged with explosives, and the humor coming from the child's innocent perspective not understanding the danger. The ... | 4b9c000df6b40d4d | 2026-05-27T02:54:59.449260+00:00 | false |
4,197 | ad24bf870ce6285d | 404 | 1bcq7up | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies all key elements of the joke: the Mafia member paying a child $10 to start his car, the child's innocent perspective not understanding the danger, and the dark irony that the mobster was using the child as a disposable bomb tester. The model even correctly notes the 'nice guy though' fram... | 52b3fd02a1b032ad | 2026-06-20T00:47:07.860037+00:00 | true |
351 | ad24bf870ce6285d | 404 | 1bcq7up | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke mechanism: dark irony involving a mafia member paying a child to start his car as a disposable bomb tester in case rivals planted a car bomb. It also captures the idea that the child is being used as a human shield against an explosion. The explanation matches the ground tru... | 4b9c000df6b40d4d | 2026-05-27T02:52:22.304905+00:00 | false |
4,195 | ad24bf870ce6285d | 404 | 1bcq7up | claude-opus-4-7 | gpt-5.4-mini | correct | The model identifies the core joke correctly: a mafia member pays a child to start his car because the child is being used as a disposable bomb detector/tester in case rivals planted a car bomb. It also captures the dark irony and the mob context. Although it adds extra framing about a painting and nostalgic storytelli... | 52b3fd02a1b032ad | 2026-06-20T00:47:07.197746+00:00 | true |
2,783 | ad24bf870ce6285d | 404 | 1bcq7up | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the core joke: a mafia member paying a child $10 to start his car as a way to test for a potential car bomb, using the child as a disposable human bomb-detector. It captures the dark irony and the innocent perspective of the child. This matches the ground truth completely. | 52b3fd02a1b032ad | 2026-06-20T00:33:30.661019+00:00 | true |
670 | ad24bf870ce6285d | 370 | 1bm5k9k | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies all key elements: the Epson printer behavior of refusing to print even black-and-white when colored ink cartridges are empty, the specific example of yellow ink being demanded unnecessarily, and how the comic translates this to a painting company scenario. The model also correctly identif... | 4b9c000df6b40d4d | 2026-05-27T02:54:31.031453+00:00 | false |
4,069 | ad24bf870ce6285d | 370 | 1bm5k9k | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke: Epson printers refuse to print even grayscale/B&W if any ink cartridge (like yellow) is empty, and this absurd logic is applied to a painting company scenario. The model specifically names Epson, explains the ink cartridge issue, and correctly identifies how the painting sc... | 52b3fd02a1b032ad | 2026-06-20T00:46:30.754092+00:00 | true |
286 | ad24bf870ce6285d | 370 | 1bm5k9k | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke mechanism: it is applying Epson printer logic to a painting company, where unnecessary yellow ink/paint is demanded despite only needing gray/black. It also captures the absurdity of Epson printers refusing to print grayscale when color ink is missing. This matches the groun... | 4b9c000df6b40d4d | 2026-05-27T02:52:10.569149+00:00 | false |
4,064 | ad24bf870ce6285d | 370 | 1bm5k9k | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke: Epson printers refusing to print black-and-white/grayscale unless all color inks, including yellow, are available, and the comic extends that absurd logic to painters refusing to make a gray house without yellow paint. It also captures the specific Epson reference and the n... | 52b3fd02a1b032ad | 2026-06-20T00:46:29.483265+00:00 | true |
2,681 | ad24bf870ce6285d | 370 | 1bm5k9k | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the core joke: Epson printers refuse to print grayscale/B&W unless all color cartridges (including yellow) are available, and the meme applies this absurd logic to a painting company painting a gray house. The model mentions Epson specifically, the yellow ink/paint demand, and the subvert... | 52b3fd02a1b032ad | 2026-06-20T00:32:09.688939+00:00 | true |
697 | ad24bf870ce6285d | 382 | 1cc1lph | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke: Benjamin Franklin was a notorious womanizer/horndog, so rather than being 'horrified' by modern America (as conservative politicians claim the Founding Fathers would be), he would be thrilled by internet pornography (xHamster). The model correctly identifies Franklin's hist... | 4b9c000df6b40d4d | 2026-05-27T02:54:43.146576+00:00 | false |
4,125 | ad24bf870ce6285d | 382 | 1cc1lph | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model's explanation correctly identifies the core joke: Ben Franklin was a known womanizer/lover of indulgence, and the joke plays on the idea that he wouldn't be 'horrified' by modern America but rather delighted by something like a porn website. The model gets the essential mechanism - using Franklin's reputation... | 52b3fd02a1b032ad | 2026-06-20T00:46:47.379143+00:00 | true |
310 | ad24bf870ce6285d | 382 | 1cc1lph | claude-opus-4-7 | gpt-5.4-mini | incorrect | The model captures part of the joke: it correctly notes that Benjamin Franklin was historically a womanizer/horndog and that the meme contrasts his serious Founding Father image with sexual indulgence. However, the ground truth’s specific humor is about Franklin’s notorious womanizing, possible illegitimate children, a... | 4b9c000df6b40d4d | 2026-05-27T02:52:14.872196+00:00 | false |
4,113 | ad24bf870ce6285d | 382 | 1cc1lph | claude-opus-4-7 | gpt-5.4-mini | incorrect | The model identifies the key reference to Benjamin Franklin and correctly notes the joke about him being a womanizer/horndog rather than a purely serious Founding Father. It also captures the contrast between Franklin's historical image and a sexualized modern reference. However, the ground truth specifically centers o... | 52b3fd02a1b032ad | 2026-06-20T00:46:43.283292+00:00 | true |
2,717 | ad24bf870ce6285d | 382 | 1cc1lph | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the core joke: Benjamin Franklin's reputation as a womanizer/lover of indulgence being contrasted with his image as a solemn Founding Father. The ground truth focuses on Franklin's womanizing, affairs, illegitimate children, and use of sexual charm in France. The model captures this same ... | 52b3fd02a1b032ad | 2026-06-20T00:32:39.351985+00:00 | true |
653 | ad24bf870ce6285d | 363 | 1cqkl76 | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies all key elements: Ea-Nasir the ancient Mesopotamian/Babylonian copper merchant, the complaint tablet from Nanni dated to 1750 BCE, the concept of the 'third death' being when your name is last spoken, and the irony that Ea-Nasir is remembered because of his preserved complaint/hate mail. ... | 4b9c000df6b40d4d | 2026-05-27T02:54:25.167258+00:00 | false |
4,039 | ad24bf870ce6285d | 363 | 1cqkl76 | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies all key elements: Ea-Nasir the ancient Mesopotamian/Babylonian copper merchant, the complaint tablet from Nanni (~1750 BCE), the internet meme status, and the ironic twist that Ea-Nasir is remembered because he preserved his own hate mail/complaint. The model also correctly identifies the... | 52b3fd02a1b032ad | 2026-06-20T00:46:23.016369+00:00 | true |
271 | ad24bf870ce6285d | 363 | 1cqkl76 | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the meme as being about Ea-Nasir, the ancient Mesopotamian/Babylonian copper merchant, and explains the key historical reference: the complaint tablet from Nanni about poor-quality copper. It also captures the irony that Ea-Nasir is remembered because of this preserved complaint, which ma... | 4b9c000df6b40d4d | 2026-05-27T02:52:08.460904+00:00 | false |
4,033 | ad24bf870ce6285d | 363 | 1cqkl76 | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies Ea-Nasir, the ancient Mesopotamian/Babylonian copper merchant, and the complaint tablet from Nanni about poor-quality copper. It also captures the core irony that Ea-Nasir is remembered because of an ancient customer complaint and that this functions like an absurdly old bad review. Altho... | 52b3fd02a1b032ad | 2026-06-20T00:46:21.638531+00:00 | true |
2,661 | ad24bf870ce6285d | 363 | 1cqkl76 | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies Ea-Nasir, the ancient Mesopotamian/Babylonian copper merchant from around 1750 BCE, the complaint tablet from Nanni about substandard copper, and the irony that Ea-Nasir is remembered because of this preserved complaint. It also correctly frames the humor as an absurdly ancient bad custom... | 52b3fd02a1b032ad | 2026-06-20T00:31:49.034859+00:00 | true |
678 | ad24bf870ce6285d | 372 | 1cycae1 | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core chain of events: 9/11 → MCR → Twilight → Fifty Shades of Grey → Ellen's downfall. It correctly identifies Gerard Way/MCR, Twilight/Stephenie Meyer, Fifty Shades of Grey, and even mentions Dakota Johnson and the birthday party incident as the connection to Ellen's downfall. The mo... | 4b9c000df6b40d4d | 2026-05-27T02:54:34.479804+00:00 | false |
4,088 | ad24bf870ce6285d | 372 | 1cycae1 | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies all the key elements of the meme chain: 9/11 → MCR (Gerard Way's inspiration) → Twilight → 50 Shades of Grey → Dakota Johnson's Ellen birthday party incident → Ellen's downfall. The model correctly identifies the Dakota Johnson birthday party incident as a viral moment contributing to Ell... | 52b3fd02a1b032ad | 2026-06-20T00:46:36.901132+00:00 | true |
290 | ad24bf870ce6285d | 372 | 1cycae1 | claude-opus-4-7 | gpt-5.4-mini | incorrect | The model captures the broad absurd chain and correctly identifies several key links: 9/11, My Chemical Romance, Twilight, Fifty Shades of Grey, Dakota Johnson, and Ellen’s downfall via the birthday-party interview. However, it misses an important specific detail from the ground truth: the causal chain is framed as Ger... | 4b9c000df6b40d4d | 2026-05-27T02:52:11.548344+00:00 | false |
4,070 | ad24bf870ce6285d | 372 | 1cycae1 | claude-opus-4-7 | gpt-5.4-mini | incorrect | The model captures the broad absurd chain from 9/11 to My Chemical Romance to Twilight to Fifty Shades of Grey to Dakota Johnson and then to Ellen's downfall. It also mentions the Ellen birthday-party interview/incident and the toxic workplace allegations that damaged Ellen's show. However, the ground truth specificall... | 52b3fd02a1b032ad | 2026-06-20T00:46:31.317314+00:00 | true |
2,687 | ad24bf870ce6285d | 372 | 1cycae1 | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the full absurd causal chain: 9/11 → My Chemical Romance → Twilight → Fifty Shades of Grey → Dakota Johnson → Ellen DeGeneres birthday party incident → Ellen's downfall. It names all the specific people, works, and events mentioned in the ground truth (Gerard Way is implied via MCR, Steph... | 52b3fd02a1b032ad | 2026-06-20T00:32:14.938959+00:00 | true |
760 | ad24bf870ce6285d | 416 | 1dl75c6 | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke mechanism: pizza orders near the Pentagon/White House spiking as an indicator of military/political crises. The ground truth mentions the 'Pizza Meter' concept, while the model calls it the 'Pentagon Pizza Index,' but both refer to the same phenomenon. The model correctly ex... | 4b9c000df6b40d4d | 2026-05-27T02:55:10.932486+00:00 | false |
4,263 | ad24bf870ce6285d | 416 | 1dl75c6 | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke about pizza deliveries near the Pentagon/White House serving as an early-warning signal for major military or political crises. The model even provides more detail than the ground truth, mentioning the 'Pentagon Pizza Index' concept, historical examples, and the visual eleme... | 52b3fd02a1b032ad | 2026-06-20T00:47:26.140901+00:00 | true |
381 | ad24bf870ce6285d | 416 | 1dl75c6 | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke mechanism: the pizza-order spike as an informal warning sign that serious Pentagon/White House crisis work is happening, making a mundane thing an absurd early indicator of major events. It also captures the humor of contradicting reassurance about war with evidence of late-... | 4b9c000df6b40d4d | 2026-05-27T02:52:27.889158+00:00 | false |
4,250 | ad24bf870ce6285d | 416 | 1dl75c6 | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke: the Pentagon/White House pizza-order spike as an informal warning sign that staff are working late on a serious crisis. It also captures the contrast between a mundane pizza surge and major geopolitical/military events. The specific reference to the 'Pentagon Pizza Index' m... | 52b3fd02a1b032ad | 2026-06-20T00:47:23.205938+00:00 | true |
2,818 | ad24bf870ce6285d | 416 | 1dl75c6 | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the core joke: the 'Pizza Meter'/'Pentagon Pizza Index' phenomenon where spikes in pizza deliveries near the Pentagon signal staffers working late on a crisis. It matches the ground truth's explanation that a mundane indicator (pizza orders) becomes an absurd early-warning signal for seri... | 52b3fd02a1b032ad | 2026-06-20T00:33:55.932323+00:00 | true |
672 | ad24bf870ce6285d | 371 | 1ejh81c | claude-opus-4-7 | claude-sonnet-4-6 | incorrect | The ground truth explains that the joke is specifically about the Wii Remote's third LED being lit, indicating it's 'Player 3,' which implies there are already two other controllers/people in the house before the husband arrived home — making him the third wheel and implying his wife is cheating. The core joke mechanis... | 4b9c000df6b40d4d | 2026-05-27T02:54:33.466605+00:00 | false |
4,094 | ad24bf870ce6285d | 371 | 1ejh81c | claude-opus-4-7 | claude-sonnet-4-6 | incorrect | The ground truth explains that the joke is about the Wii Remote's LED indicator showing it's player 3 (third LED lit), which implies there are already two other controllers/people in the house, suggesting the wife is cheating and making the husband the 'third wheel' or third player. The model completely misses this spe... | 52b3fd02a1b032ad | 2026-06-20T00:46:38.588714+00:00 | true |
288 | ad24bf870ce6285d | 371 | 1ejh81c | claude-opus-4-7 | gpt-5.4-mini | incorrect | The model does identify a cheating-wife style joke, but it misses the actual mechanism in the ground truth. The ground truth says the Wii Remote’s third LED indicates player 3, implying two other controllers/people are already present, so the husband is the third person and infers his wife is cheating. The model instea... | 4b9c000df6b40d4d | 2026-05-27T02:52:11.062883+00:00 | false |
4,067 | ad24bf870ce6285d | 371 | 1ejh81c | claude-opus-4-7 | gpt-5.4-mini | incorrect | The model does not match the ground truth. The ground truth says the joke is about the Wii Remote’s third LED indicating player 3, implying two other controllers/people are already present, so the husband is the third person and therefore his wife is cheating. The model instead invents a different joke about a Wii Remo... | 52b3fd02a1b032ad | 2026-06-20T00:46:30.265902+00:00 | true |
2,683 | ad24bf870ce6285d | 371 | 1ejh81c | claude-opus-4-7 | z-ai/glm-5.2 | incorrect | The model completely misses the actual joke identified in the ground truth. The ground truth explains that the Wii Remote's third LED indicates it is player 3, implying two other people/controllers are already in the house, thus suggesting the wife is cheating and the husband is the 'third person.' The model, however, ... | 52b3fd02a1b032ad | 2026-06-20T00:32:11.178254+00:00 | true |
740 | ad24bf870ce6285d | 406 | 1eu8ulz | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke: students making up 'grandma died' excuses to skip early 8 AM classes. The model also correctly explains the mechanism — moving the class to 3 PM stops the fake deaths, so the professor 'saved lives.' This matches the ground truth's explanation about the overused excuse bein... | 4b9c000df6b40d4d | 2026-05-27T02:55:02.360602+00:00 | false |
4,207 | ad24bf870ce6285d | 406 | 1eu8ulz | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke: students use 'my grandma died' as a fake excuse to skip 8 AM classes, and the professor humorously treats this as if it were real by claiming moving the class to 3 PM 'saved lives' (since no more grandparents died). The model captures all the key elements: the overused fake... | 52b3fd02a1b032ad | 2026-06-20T00:47:10.849093+00:00 | true |
355 | ad24bf870ce6285d | 406 | 1eu8ulz | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke mechanism: students using the overused 'my grandma/grandpa died' excuse to avoid an 8 AM class, and the humor coming from treating that obviously fake excuse as a normal pattern. It also captures why the punchline works by implying that moving the class later 'saved lives' b... | 4b9c000df6b40d4d | 2026-05-27T02:52:23.114380+00:00 | false |
4,200 | ad24bf870ce6285d | 406 | 1eu8ulz | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke: students use the overused fake excuse that a grandma/grandpa died to avoid an 8 AM class, and the humor comes from treating that obviously bogus pattern as if it were real. It also captures the contrast that moving the class later results in no more 'deaths,' implying the e... | 52b3fd02a1b032ad | 2026-06-20T00:47:08.648366+00:00 | true |
2,789 | ad24bf870ce6285d | 406 | 1eu8ulz | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the core joke: students using the overused 'my grandma died' excuse to skip 8 AM classes, and the humor in treating this obviously fake excuse as a real pattern. The model also adds additional context about the professor's deadpan response and the punchline about 'saving lives,' which is ... | 52b3fd02a1b032ad | 2026-06-20T00:33:34.156222+00:00 | true |
733 | ad24bf870ce6285d | 402 | 1g8raiu | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core reference: Dale Gribble from King of the Hill and his 'pocket sand' gag. It mentions Dale Gribble by name, references King of the Hill specifically, and explains that pocket sand is his signature self-defense move where he throws sand in people's faces. The model understands WHY ... | 4b9c000df6b40d4d | 2026-05-27T02:54:59.220623+00:00 | false |
4,191 | ad24bf870ce6285d | 402 | 1g8raiu | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core reference: Dale Gribble from King of the Hill and his 'pocket sand' gag. It correctly explains that Dale throws sand in people's faces as a self-defense move, and correctly captures the humor of treating pocket sand as an essential item. The model even mentions the visual presenc... | 52b3fd02a1b032ad | 2026-06-20T00:47:06.027596+00:00 | true |
348 | ad24bf870ce6285d | 402 | 1g8raiu | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the specific reference to Dale Gribble from King of the Hill and explains the 'pocket sand' gag as a ridiculous self-defense move involving throwing sand into someone's eyes. It also captures the humor that pocket sand is treated like an essential item alongside wallet/keys/phone, which m... | 4b9c000df6b40d4d | 2026-05-27T02:52:21.621523+00:00 | false |
4,185 | ad24bf870ce6285d | 402 | 1g8raiu | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the specific reference to Dale Gribble from King of the Hill and the 'pocket sand' gag, including the absurd self-defense move of throwing sand in someone's eyes. It also captures the core joke that pocket sand is treated as an essential item alongside wallet/keys/phone, which matches the... | 52b3fd02a1b032ad | 2026-06-20T00:47:04.558293+00:00 | true |
2,776 | ad24bf870ce6285d | 402 | 1g8raiu | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the specific reference to Dale Gribble from King of the Hill and his 'pocket sand' gag. It explains the humor as the absurdity of treating pocket sand as an essential everyday carry item, which aligns with the ground truth's emphasis on the ridiculousness of pocket sand as a signature sel... | 52b3fd02a1b032ad | 2026-06-20T00:33:24.945202+00:00 | true |
728 | ad24bf870ce6285d | 399 | 1gqh1c5 | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke mechanism: using professional kitchen safety calls (like 'behind,' 'corner') in an ordinary, low-stakes office break room setting while making Cup Noodles. The model understands the absurdity of applying high-pressure kitchen mindset to a mundane task, which is the exact jok... | 4b9c000df6b40d4d | 2026-05-27T02:54:57.093681+00:00 | false |
4,196 | ad24bf870ce6285d | 399 | 1gqh1c5 | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke: using professional kitchen safety calls ('behind,' 'corner') in a mundane office break room setting while making Cup Noodles. The model captures the key elements: the kitchen call-outs, the absurd mismatch between high-pressure kitchen behavior and low-stakes office environ... | 52b3fd02a1b032ad | 2026-06-20T00:47:07.406890+00:00 | true |
347 | ad24bf870ce6285d | 399 | 1gqh1c5 | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke mechanism: professional kitchen safety calls like “corner” and “behind” being used in an ordinary office break room while making Cup Noodles, creating humor through absurd over-seriousness and mismatch of context. However, the ground truth also specifically mentions calls li... | 4b9c000df6b40d4d | 2026-05-27T02:52:21.575021+00:00 | false |
4,187 | ad24bf870ce6285d | 399 | 1gqh1c5 | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke: professional kitchen safety calls like "Corner!" and "Behind!" being used in an ordinary office break room while making Cup Noodles. This matches the ground truth’s point about absurd over-seriousness and the mismatch between high-pressure kitchen language and a low-stakes ... | 52b3fd02a1b032ad | 2026-06-20T00:47:04.946586+00:00 | true |
2,767 | ad24bf870ce6285d | 399 | 1gqh1c5 | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the kitchen safety calls ('Corner!', 'Behind!'), the contrast between professional kitchen culture and the mundane office break room setting, and the specific detail of making cup noodles. It matches the ground truth's core points: the absurd over-seriousness, the mismatch between high-pr... | 52b3fd02a1b032ad | 2026-06-20T00:33:17.337497+00:00 | true |
681 | ad24bf870ce6285d | 375 | 1h1iscl | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies: (1) the '#WomenInMaleFields' TikTok trend as the context, (2) the core joke mechanism - mocking the classic male 'I could have gone pro if not for my knee injury' cliché, and (3) the gender-flip humor where this typically male behavior is applied to a female context. The ground truth foc... | 4b9c000df6b40d4d | 2026-05-27T02:54:37.351713+00:00 | false |
4,086 | ad24bf870ce6285d | 375 | 1h1iscl | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core elements of the meme: it's a gender-flipped '#Womeninmalefields' TikTok trend joke that mocks the male cliché of claiming they 'could have gone pro' if not for a career-ending knee injury. The model gets the specific TikTok trend reference, the knee injury excuse trope, and the g... | 52b3fd02a1b032ad | 2026-06-20T00:46:36.662008+00:00 | true |
294 | ad24bf870ce6285d | 375 | 1h1iscl | claude-opus-4-7 | gpt-5.4-mini | incorrect | The model captures the broad structure of the meme: a gender-flipped 'women in male fields' joke and the familiar 'I could have gone pro if not for a knee injury' excuse. However, the ground truth specifically says the joke is about men claiming they would have gone pro in football/soccer or rugby if not for a career-e... | 4b9c000df6b40d4d | 2026-05-27T02:52:12.147777+00:00 | false |
4,078 | ad24bf870ce6285d | 375 | 1h1iscl | claude-opus-4-7 | gpt-5.4-mini | incorrect | The model correctly identifies the core joke as a gender-flipped version of the familiar 'I could have gone pro if not for a knee injury' excuse. It also captures the self-mythologizing, hard-to-disprove bragging about unrealized athletic potential. However, it adds an incorrect specific twist by saying the woman claim... | 52b3fd02a1b032ad | 2026-06-20T00:46:33.624111+00:00 | true |
2,696 | ad24bf870ce6285d | 375 | 1h1iscl | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the '#Womeninmalefields' trend, the cliché of men claiming they 'could have gone pro' if not for a career-ending injury, and the gender-flipped humor of applying this male bragging pattern to a feminine aspiration. It captures the same core joke as the ground truth: mocking the familiar, ... | 52b3fd02a1b032ad | 2026-06-20T00:32:22.395300+00:00 | true |
706 | ad24bf870ce6285d | 381 | 1hb73xj | claude-opus-4-7 | claude-sonnet-4-6 | incorrect | The ground truth specifically refers to the person who snitched on the CEO shooter (Luigi Mangione / UnitedHealthcare killer case) expecting the $50k reward but being denied due to a technicality (his tip led to detention/questioning rather than the actual arrest). The model gives a generic explanation about powerful e... | 4b9c000df6b40d4d | 2026-05-27T02:54:48.667548+00:00 | false |
4,106 | ad24bf870ce6285d | 381 | 1hb73xj | claude-opus-4-7 | claude-sonnet-4-6 | incorrect | The ground truth specifically identifies the meme as being about the person who snitched on the UHC CEO shooter (Luigi Mangione case), who expected a $50k reward but gets denied it due to a technicality (his tip led to detention/questioning rather than the actual arrest). The model gives a generic interpretation about ... | 52b3fd02a1b032ad | 2026-06-20T00:46:41.919450+00:00 | true |
304 | ad24bf870ce6285d | 381 | 1hb73xj | claude-opus-4-7 | gpt-5.4-mini | incorrect | The model did not identify the specific meme reference or the actual joke mechanism from the ground truth. The ground truth is about a very specific situation: someone who snitched on the CEO shooter / UHC killer expects a $50k reward, but is denied because the tip only led to detention/questioning, not the actual arre... | 4b9c000df6b40d4d | 2026-05-27T02:52:14.028681+00:00 | false |
4,099 | ad24bf870ce6285d | 381 | 1hb73xj | claude-opus-4-7 | gpt-5.4-mini | incorrect | The ground truth is very specific: the meme is about the person who snitched on the CEO shooter / UHC killer expecting a $50k reward, but being denied because his tip only led to detention/questioning, not the actual arrest. The model instead describes a completely different meme template ('Laughing Men in Suits') and ... | 52b3fd02a1b032ad | 2026-06-20T00:46:39.457183+00:00 | true |
2,710 | ad24bf870ce6285d | 381 | 1hb73xj | claude-opus-4-7 | z-ai/glm-5.2 | incorrect | The ground truth specifically identifies the joke as being about the person who snitched on the CEO shooter / UHC killer expecting a $50k reward, but being denied due to a technicality (tip led to detention/questioning rather than actual arrest). The model's explanation is entirely generic, talking about 'reward money'... | 52b3fd02a1b032ad | 2026-06-20T00:32:34.284680+00:00 | true |
719 | ad24bf870ce6285d | 395 | 1hfggtz | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke: the difference between American drywall (soft, easy to punch through) and European brick/concrete walls (hard, will break your hand). The model identifies the same joke mechanism - the contrast between the two types of walls and the different outcomes of punching them. The ... | 4b9c000df6b40d4d | 2026-05-27T02:54:53.380562+00:00 | false |
4,225 | ad24bf870ce6285d | 395 | 1hfggtz | claude-opus-4-7 | claude-sonnet-4-6 | correct | The model correctly identifies the core joke: the contrast between punching American drywall (easy to punch through, soft) versus European walls made of brick/concrete (hard, hurts your hand). The model correctly identifies the setup with both panels, the cultural/architectural difference, and the humor derived from th... | 52b3fd02a1b032ad | 2026-06-20T00:47:15.609549+00:00 | true |
364 | ad24bf870ce6285d | 395 | 1hfggtz | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke mechanism: a contrast between punching soft American drywall and punching a much harder wall that would injure your hand. This matches the ground truth's subverted expectation that both are 'walls' but one is easy to damage while the other is not. The explanation also captur... | 4b9c000df6b40d4d | 2026-05-27T02:52:24.717187+00:00 | false |
4,218 | ad24bf870ce6285d | 395 | 1hfggtz | claude-opus-4-7 | gpt-5.4-mini | correct | The model correctly identifies the core joke: a contrast between punching American drywall, which is soft and easy to damage, and punching a real brick/concrete wall, which would injure the puncher. This matches the ground truth's setup, subverted expectation, and the humor from one wall being much softer than the othe... | 52b3fd02a1b032ad | 2026-06-20T00:47:13.640927+00:00 | true |
2,755 | ad24bf870ce6285d | 395 | 1hfggtz | claude-opus-4-7 | z-ai/glm-5.2 | correct | The model correctly identifies the core joke: the contrast between American drywall (soft, easy to punch through) and European brick/concrete walls (hard, will injure your hand). It matches the ground truth's explanation of the subverted expectation and the difference in material hardness. The model also correctly desc... | 52b3fd02a1b032ad | 2026-06-20T00:33:09.527954+00:00 | true |
689 | ad24bf870ce6285d | 380 | 1hg2bou | claude-opus-4-7 | claude-sonnet-4-6 | incorrect | The ground truth specifies that the meme jokes about Kermit having an AR-15 on his back and a coat hanger bent into a makeshift drop-in auto sear (DIAS) - a specific illegal firearm modification to convert a semi-automatic rifle to full auto. The model's explanation recognizes the general concept of the coat hanger bei... | 4b9c000df6b40d4d | 2026-05-27T02:54:39.641141+00:00 | false |
4,111 | ad24bf870ce6285d | 380 | 1hg2bou | claude-opus-4-7 | claude-sonnet-4-6 | incorrect | The ground truth explains that the meme's joke is specifically about using a coat hanger bent into a makeshift drop-in auto sear (DIAS) to illegally convert an AR-15 from semi-automatic to full auto. This is a very specific gun-related joke about illegal weapon modification.
The model's explanation correctly identifie... | 52b3fd02a1b032ad | 2026-06-20T00:46:42.868197+00:00 | true |
303 | ad24bf870ce6285d | 380 | 1hg2bou | claude-opus-4-7 | gpt-5.4-mini | incorrect | The model correctly identifies the general 'Nervous Kermit' guilt/anxiety reaction and the idea that bent hangers imply some shady or illegal activity. However, it misses the specific ground-truth joke: the hanger is not just a vague sinister use, but a coat hanger bent into a makeshift drop-in auto sear used to illega... | 4b9c000df6b40d4d | 2026-05-27T02:52:13.809024+00:00 | false |
4,096 | ad24bf870ce6285d | 380 | 1hg2bou | claude-opus-4-7 | gpt-5.4-mini | incorrect | The model correctly identifies the Nervous Kermit image and the general setup of someone being panicked about incriminating hanger-related activity. However, it misses the specific joke in the ground truth: the hanger is not just a vague shady use, but a coat hanger bent into a makeshift drop-in auto sear, and Kermit h... | 52b3fd02a1b032ad | 2026-06-20T00:46:38.842807+00:00 | true |
2,711 | ad24bf870ce6285d | 380 | 1hg2bou | claude-opus-4-7 | z-ai/glm-5.2 | incorrect | The ground truth specifically identifies that the meme involves Kermit having an AR-15 with a coat hanger bent into a makeshift drop-in auto sear, implying an illegal conversion from semi-automatic to full auto. The model's explanation only generically suggests the bent hangers were used for 'something illegal or sinis... | 52b3fd02a1b032ad | 2026-06-20T00:32:35.536795+00:00 | true |
End of preview. Expand in Data Studio
BasedBench: basedBench-519-2026-07
BasedBench is a VLM meme-understanding benchmark. This snapshot contains 519 human-validated memes with ground-truth explanations derived from Reddit comment consensus.
Task Definition
The task is to determine whether a model gets the joke in a meme. A correct prediction identifies the relevant people, events, meme formats, media, phrases, visual details, or cultural references, then reconstructs the intended setup, implication, contrast, inversion, irony, wordplay, or other mechanism a viewer must notice to understand the meme.
This benchmark does not test whether a model can produce a psychological or aesthetic theory of why something is funny.
Leaderboard
| Model | Correct | Total | Accuracy |
|---|---|---|---|
| claude-opus-4-8 | 312 | 519 | 60.1% |
| google/gemini-3.1-pro-preview | 442 | 519 | 85.2% |
| gpt-5.5 | 419 | 519 | 80.7% |
| muse-spark-1.1 | 395 | 519 | 76.1% |
| x-ai/grok-4.3 | 254 | 519 | 48.9% |
Dataset Contents
The public artifact includes:
- Reddit post IDs, titles, and subreddit names.
- Meme images used as benchmark stimuli.
- Human-validated ground-truth explanations.
- Every successful model prediction retained by BasedBench for this snapshot.
- Every judge verdict and reasoning record for those predictions, including superseded rejudgments.
- Derived consensus fields and leaderboard totals.
Raw Reddit comments, Reddit authors, review metadata, reviewer notes, consensus source comment IDs, local file paths, API request metadata, internal prompts, raw LLM responses, and LLM call logs are intentionally omitted.
Usage
from datasets import load_dataset
memes = load_dataset("montagovian/basedBench", "memes")
predictions = load_dataset("montagovian/basedBench", "predictions")
judgments = load_dataset("montagovian/basedBench", "judgments")
leaderboard = load_dataset("montagovian/basedBench", "leaderboard")
Dataset Structure
The dataset uses normalized long-form tables joined by snapshot_id, post_id,
and prediction_id:
memeshas one row per meme in the snapshot, with the image, title, subreddit, and human-validated ground truth.predictionshas one row per successful model prediction. It includes the target model, prediction text, dataset and prompt versions where available, latency, token count, timestamp, and derived consensus vote fields.judgmentshas one row per judgment attempt. It includes the target and judge models, verdict, reasoning, judge prompt ID, timestamp, and anis_latestmarker. Historical rejudgments remain present.leaderboardhas one derived row per target model, including score coverage and judge agreement statistics.
Failed API calls and operational error messages are not benchmark predictions
and are omitted from the public tables. The current BasedBench database retains
one successful prediction per (post_id, model_id) pair; historical successful
prediction reruns that were never retained by the database cannot be exported.
Methodology
Ground-truth explanations are extracted from Reddit comments via LLM consensus detection. A candidate ground truth must reflect at least three substantive comments agreeing on the same specific explanation, and a human reviewer must validate the meme before it enters a release snapshot.
Predictor models receive only the meme image. They do not receive the Reddit title, comments, subreddit, ground truth, web search, or external tools.
Each model prediction is scored by an LLM judge ensemble using strict criteria:
the judge asks whether the model recovered the same joke as the ground truth.
For derived consensus fields and leaderboard accuracy, only the latest judgment
from each (prediction_id, judge_model) pair is counted. A prediction receives
a consensus verdict when at least two judges cast the same verdict and that
verdict has a strict majority; otherwise it is omitted from the leaderboard
denominator. All individual and historical judgments remain available in the
judgments config.
License and Rights
This dataset has mixed rights status, so the machine-readable Hugging Face
license is other.
Materials created and controlled by the BasedBench maintainers are released under the MIT License. This includes the benchmark code, dataset schema, export format, evaluation prompts where applicable, benchmark-specific metadata, leaderboard tables, judge verdicts and reasoning, and other maintainer-authored documentation and annotations to the extent the maintainers own or control those materials.
Meme images, Reddit post titles, subreddit names, post IDs, cultural references, logos, characters, screenshots, and other source artifacts may be owned by third parties. The BasedBench maintainers do not claim ownership of those underlying third-party materials. The MIT License for this repository does not apply to those third-party materials; all such rights remain with their respective owners.
The meme images and limited source metadata are included under a fair-use rationale for research, criticism, commentary, and benchmark evaluation. The use is transformative: the images are used as individual test stimuli for measuring whether vision-language models understand the intended joke, not as a substitute for the original posts or images. The dataset uses only the material needed to support that benchmark task, omits raw comment text and authors, and does not serve as a general-purpose meme archive or replacement market for the source works.
Users are responsible for determining whether their downstream use of any third-party material is permitted by law or by the relevant rights holder. If you believe specific material should not be included, please contact the dataset maintainers through the Hugging Face repository or the project repository.
Intended Use
This dataset is intended for research and evaluation of multimodal model understanding, especially whether models can connect visual details, text, cultural references, and joke structure.
It is not intended for training models to impersonate Reddit users, reconstruct deleted discussions, identify commenters, or build a general meme redistribution corpus.
Snapshot
- Snapshot name:
basedBench-519-2026-07 - Snapshot ID:
ad24bf870ce6285d - Created:
2026-07-15T01:49:13.506333+00:00 - Memes: 519
- Successful predictions: 2785
- Judgment records: 10660
- Downloads last month
- 103