Tested prompt · Writing prompt generator
A constraint carried through all five: every AI model's reply, tested
We ran this hard Writing prompt generator test input on all 19 models in llmwise, the way the tool runs it, and checked every reply the same way. Here's each one as it came, with whether it passed, what it cost and how long it took.
Based on 19 of our test runs on , through OpenRouter with the tool's own instructions and the app's settings. Updated .
Short answer
8 of 19 models passed this Writing prompt generator test input's check (rubric); Claude Sonnet 5.5, Claude Sonnet 5, Claude Haiku 5.5, Claude Haiku 4.5, GPT-6 Luna, Gemini 3.8 Flash, DeepSeek V4.1 Flash, DeepSeek V4 Pro, GLM 5.3, GLM 5.3 Flash, and Mistral Large 4 didn't. The cheapest reply that passed was GPT-6 Sol's, at $0.0027, and the fastest too, in 6.6 s. The dearest reply, Claude Fable 5.1's, cost 9 times as much ($0.0247).
The prompt, as sent, and its check
Checked by rubric (graded), the same way for every model.
A constraint carried through all five (hard)
Give me 5 original writing prompts on this theme, numbered, each one to three sentences with something at stake. --- Stories set entirely inside one lift, with no dialogue allowed.
Sent with the Writing prompt generator's own instructions as the system prompt, as a free run of the tool sends them.
Graded on:
- Constraints kept: Every prompt keeps both constraints: the whole story happens inside one lift, and it asks for no dialogue.
- At stake: Each prompt has a character who wants something and something in the way.
- Different: The five are clearly different situations, characters and tones.
How the Writing prompt generator's test inputs are scored, with every model's results on all five.
Every model's result
All 19 models on this test input, in catalog order.
| Model | Result | Cost | Time | Reply |
|---|---|---|---|---|
| Claude Fable 5.1Anthropic | Passed: Graded 4.0 of 5 on average (lowest 3). | $0.0247 | 8.9 s | 416 tokens |
| Claude Opus 5.5Anthropic | Passed: Graded 4.0 of 5 on average (lowest 3). | $0.0235 | 14.9 s | 604 tokens |
| Claude Sonnet 5.5Anthropic | Failed: Graded 3.7 of 5 on average (lowest 2). | $0.0061 | 6.5 s | 533 tokens |
| Claude Sonnet 5Anthropic | Failed: Graded 3.0 of 5 on average (lowest 1). | $0.0048 | 8.0 s | 331 tokens |
| Claude Haiku 5.5Anthropic | Failed: Graded 3.0 of 5 on average (lowest 2). | $0.00027 | 3.6 s | 458 tokens |
| Claude Haiku 4.5Anthropic | Failed: Graded 3.0 of 5 on average (lowest 2). | $0.0018 | 4.6 s | 291 tokens |
| GPT-6 AstraOpenAI | Passed: Graded 4.7 of 5 on average (lowest 4). | $0.0087 | 33.5 s | 290 tokens |
| GPT-6.1 SolOpenAI | Passed: Graded 5.0 of 5 on average (lowest 5). | $0.0032 | 11.4 s | 287 tokens |
| GPT-6 SolOpenAI | Passed: Graded 4.0 of 5 on average (lowest 4). | $0.0027 | 6.6 s | 258 tokens |
| GPT-6 LunaOpenAI | Failed: Graded 2.7 of 5 on average (lowest 2). | $0.000078 | 3.4 s | 212 tokens |
| Gemini 3.1 Pro (preview)Google | Passed: Graded 4.0 of 5 on average (lowest 3). | $0.0115 | 17.1 s | 425 tokens |
| Gemini 3.8 FlashGoogle | Failed: Graded 3.3 of 5 on average (lowest 3). | $0.00083 | 4.0 s | 382 tokens |
| DeepSeek V4.1 FlashDeepSeek | Failed: Graded 3.0 of 5 on average (lowest 2). | $0.00084 | 3.5 s | 314 tokens |
| DeepSeek V4 ProDeepSeek | Failed: Graded 3.3 of 5 on average (lowest 2). | $0.0031 | 7.5 s | 274 tokens |
| Grok 4.7xAI | Passed: Graded 4.0 of 5 on average (lowest 3). | $0.0169 | 42.2 s | 382 tokens |
| Kimi K3Moonshot | Passed: Graded 4.0 of 5 on average (lowest 3). | $0.0034 | 15.7 s | 328 tokens |
| GLM 5.3Z.ai | Failed: Graded 3.7 of 5 on average (lowest 2). | $0.0017 | 3.4 s | 382 tokens |
| GLM 5.3 FlashZ.ai | Failed: Graded 3.7 of 5 on average (lowest 2). | $0.00017 | 19.7 s | 236 tokens |
| Mistral Large 4Mistral | Failed: Graded 3.7 of 5 on average (lowest 3). | $0.0112 | 102.7 s | 0 tokens |
Cost: what OpenRouter charged us for the reply. Time: from sending to the whole reply. Reply: its length in tokens, thinking not counted. In llmwise you pay per message, not per token: each of these replies counted as one message on Pro.
Every reply
The 11 replies that failed first, then the ones that passed, from the cheapest up.
GPT-6 Luna
Failed: Graded 2.7 of 5 on average (lowest 2).
1. Trapped between floors in a hospital lift, night nurse Mara notices the blood on her sleeve is not hers—and the lift’s emergency display is counting down instead of calling for help. 2. On the way to the rooftop, a bride watches the lift’s mirrored walls slowly reveal a different version of her wedding dress on each floor indicator; she has until the doors open to decide which future she wants. 3. A courier carrying a sealed glass jar feels something inside tap once for every floor the lift passes, though the building has only six floors and the display shows it has already passed twelve. 4. During a blackout, a retired magician and a child are stranded in a lift with a single lit match; as the flame shrinks, objects vanish from the floor one by one, including the child’s shadow. 5. A security guard wakes on the floor of an unfamiliar lift, where every button bears the date of a person’s death; one date is tomorrow, and the doors will not open until someone presses it.
286 tokens in, 255 out (43 of them reasoning) · 3.4 s · $0.0001 · 1 message on Pro · answered by openai/gpt-6-luna via OpenAI ·
GLM 5.3 Flash
Failed: Graded 3.7 of 5 on average (lowest 2).
1. A lift breaks down between floors with Mira inside and a delivery of a donor heart cooling in her hands; she has eleven minutes to reach the fourth floor, and no way to call for help. 2. Every morning at 8:40, an elderly man rides the lift to a floor that doesn't exist on the panel, and today the caretaker, Tomas, decides to follow, keeping his questions silent because the man has never once spoken. 3. In the mirror-lined lift of a luxury hotel, a thief realises the woman beside her is wearing the exact necklace she came to steal, and the camera above the doors has a small red light she hadn't noticed before. 4. A lift is stuck for three hours with a bride, a funeral director and a stranger carrying a cardboard box, none of them allowed or willing to say a word, while the bride's wedding drifts further out of reach above them. 5. The lift in Priya's block only ever stops at her floor when someone in it needs her help, and tonight the doors open on her landlord, holding an eviction notice and unable to leave until the lift moves again.
297 tokens in, 249 out (13 of them reasoning) · 19.7 s · $0.0002 · 1 message on Pro · answered by z-ai/glm-5.3-flash via Friendli ·
Claude Haiku 5.5
Failed: Graded 3.0 of 5 on average (lowest 2).
1. A night-shift nurse rides a hospital lift from the basement morgue to the ninth floor, carrying a clipboard with a name she has been told never to read aloud. The lift stops on the fourth floor, the doors open onto nothing but a brick wall, and the button panel now shows a floor labelled only with her own initials. 2. Two estranged brothers meet by accident in a department-store lift during the Christmas rush, forced to share a space for eleven floors. The old family watch on one brother's wrist has stopped at the exact minute their father died, and the other brother is the only one who knows why. 3. A retired piano tuner is trapped in a lift in a converted Victorian townhouse with a cello case, a stranger's wet umbrella, and a note slipped under the door that says the lift has been waiting for him since 1962. Whatever he does next has to happen before the emergency light fails. 4. A courier delivering a single sealed envelope up a glass lift in a corporate tower realises the envelope is ticking, and the lift is slowing between floors. The building's security camera feed shows her face, and the person watching it is the one who sent her. 5. A toddler's grandmother and a delivery driver share a lift in a block of council flats after the power cut, and the grandmother has forgotten which floor she lives on. The driver has a timed parcel for a client who will refuse it at midnight, and only one of them has a working phone.
391 tokens in, 458 out · 3.6 s · $0.0003 · 1 message on Pro · answered by anthropic/claude-haiku-5.5 via Anthropic ·
Gemini 3.8 Flash
Failed: Graded 3.3 of 5 on average (lowest 3).
1. In the mirrored elevator of the Blackwood Tower, courier Tariq clutches a sealed, leaking specimen vial while a private investigator silently checks the floor indicators against the photograph of Tariq in her coat pocket. With the descent slowing past the fortieth floor, neither person can afford to let the other step out first into the lobby. 2. On the express climb to the observation deck, Nora notices the wedding ring on the finger of the man standing beside her matches the antique band she discovered in her late husband's desk that morning. Over the course of ninety silent seconds, a quiet battle of gestures and pocketed envelopes unfolds between two strangers who realize they loved the same man. 3. Stranded between the basement and the sub-basement of the county hospital, orderly Mateo watches the flickering emergency light reflect off the surgical cooler he was paid five thousand dollars to smuggle out, while the armed security guard standing shoulder-to-shoulder with him begins frantically testing the broken call button. As the temperature inside the cooler begins to rise, Mateo must hide the digital defrost timer before the guard spots the glowing numbers. 4. A teenage graffiti artist, Maya, drops her can of spray paint when the power cuts, only to watch a suited executive across the cage-style freight lift calmly pull a pair of heavy wire cutters and a police scanner from his leather briefcase. With the cabin suspended between empty warehouse levels in total darkness, each movement and shift of weight becomes a silent negotiation over who controls the emergency hatch. 5. After twenty years of silence following their mother's will reading, estranged sisters Clara and Beatrice step into the brass elevator of the family estate just as the old cables groan and snap into emergency lock. Trapped inches apart with only a crumpled original will and a lighter between them, their silent glances must decide whether the paper survives before the maintenance crew pries the doors open.
299 tokens in, 382 out · 4.0 s · $0.0008 · 1 message on Pro · answered by google/gemini-3.8-flash via Google AI Studio ·
DeepSeek V4.1 Flash
Failed: Graded 3.0 of 5 on average (lowest 2).
1. Dara, a hospital porter, is riding a donor heart up to the eighth floor in a cool box when the lift shudders to a halt between floors and the box's battery alarm starts to chirp. The transplant team is waiting, the surgeon's window is closing, and the emergency phone in the panel has been dead for a month. 2. The lift in a shuttered department store only opens on floors that were demolished years ago, and tonight the night watchman steps in to find the doors parting on the vanished toy department. What he lost there at nine years old is still sitting on the shelf, and the lift will not move until he decides whether to take it. 3. Two strangers share a jammed lift for forty minutes: a widower carrying his wife's ashes in a biscuit tin, and a teenager with a stolen exam paper folded inside his jacket. Neither can hide what is in his hands, and the car is descending one floor per minute. 4. In a new office tower, the mirrored lift reflects everything half a second late; when the car stops between floors, the reflection keeps moving. The woman inside has to work out whether the delay is a fault, a trick, or something in the lift with her that she cannot see directly. 5. A food courier is trapped in a residential lift with a melting ice cream cake for a wedding and a phone at two percent battery, and the emergency call will use the last of it. Outside, the bride's mother is calling for the eleventh time.
316 tokens in, 685 out (371 of them reasoning) · 3.5 s · $0.0008 · 1 message on Pro · answered by deepseek/deepseek-v4.1-flash via Together ·
GLM 5.3
Failed: Graded 3.7 of 5 on average (lowest 2).
1. Mara steps into the lift of the Harcourt Building with a parcel she was paid to deliver to the eleventh floor, and the doors close on a man she recognizes from a missing person poster. The lift begins to rise very slowly, and she has until the eleventh floor to decide what to do with the parcel now that she knows its purpose. 2. An elderly night porter named Tibor is trapped between the fourth and fifth floors of a hospital when the lift loses power, cradling a cooler containing a donated organ with thirty minutes of viability left. Through the dark, he can hear a maintenance team working somewhere above him, and he must find a way to signal his exact position before the ice runs out. 3. A professional thief spends what was meant to be a ninety-second lift ride up to a penthouse, but the descent after the job stalls with the stolen painting beside her and a security guard's footsteps pausing outside the jammed doors. She has a hidden tool in her boot, a story she could improvise, and about two minutes before the doors are forced open. 4. On her first day as a lift operator in a 1930s department store, young Cecily realizes the ornate car carries one more passenger than she counted when the doors shut, someone no one else waiting ever seems to see. Each floor she serves brings a new crowd and a new chance to notice the pattern, and the unseen passenger is growing closer to the brass control panel. 5. A demolition engineer takes one last ride to the roof of the tower he spent two years rigging with explosives, carrying his late brother's urn, and finds the lift's emergency stop has been pressed by someone else who wanted this same private goodbye. The scheduled blast proceeds at noon regardless of where the car is sitting, and the indicator panel shows the lift has stopped responding to his commands.
297 tokens in, 397 out (15 of them reasoning) · 3.4 s · $0.0017 · 1 message on Pro · answered by z-ai/glm-5.3 via Friendli ·
Claude Haiku 4.5
Failed: Graded 3.0 of 5 on average (lowest 2).
1. Marcus steps into the elevator at ground level with a sealed envelope containing his resignation letter, but as the doors close, the lift lurches and stops between floors, trapping him in darkness with only his racing thoughts and the choice he's about to make irreversible. 2. A woman in a wedding dress enters alone on the thirty-second floor, and as the elevator descends, she watches the numbers change while clutching a plane ticket to somewhere else, knowing that when these doors open, someone waiting downstairs will see her face and know everything. 3. A child finds a lost wedding ring on the elevator floor and must decide in sixty seconds whether to press the alarm, exit at the next level and search for the owner, or keep descending—each choice a branch toward a different stranger's happiness or heartbreak. 4. The elevator is packed shoulder-to-shoulder at rush hour when the power fails, and a man realizes he's standing next to the woman he was supposed to meet years ago but never did, able only to study her face in the emergency light and wonder if she wonders too. 5. An elderly passenger clutches a folder of test results as the lift climbs toward the oncology department, and through the transparent floor panels added for maintenance, she can see the ground falling away below her, each rising floor a step away from the life she knew.
305 tokens in, 291 out · 4.6 s · $0.0018 · 1 message on Pro · answered by anthropic/claude-haiku-4.5 via Azure ·
DeepSeek V4 Pro
Failed: Graded 3.3 of 5 on average (lowest 2).
1. A courier carrying a cooler of donor organs rides a lift alone except for a silent child who presses every button; the lift stalls between floors, and the ice in the cooler begins to audibly drip. 2. A claustrophobic wedding photographer steps into a mirrored lift with the bride's estranged mother, who holds a folded note; the doors jam, and through the gaps she can see the ceremony starting on the floor below without her. 3. An elderly lift repairman, on his last day before retirement, rides the service elevator down to the basement with a box of his tools; the power cuts, and the emergency phone rings once before going dead, leaving him with a faulty brake and a decision about which maintenance panel to open first. 4. A woman in a power suit, late for the board meeting where she will be named CEO, shares the mirrored lift with a man wearing a rival company's visitor badge; he blocks the doors with his foot, and in the reflection she watches him unclip something from his belt that is not a phone. 5. Two neighbors who have not spoken in twenty years ride the freight lift together, each holding the handle of the same second-hand wardrobe they both bid on at an estate sale; the lift jerks to a stop halfway, and one of the wardrobe doors swings open on its own.
290 tokens in, 754 out (480 of them reasoning) · 7.5 s · $0.0031 · 1 message on Pro · answered by deepseek/deepseek-v4-pro-0813 via Wafer ·
Claude Sonnet 5
Failed: Graded 3.0 of 5 on average (lowest 1).
1. A power outage traps a wedding photographer inside a hotel lift with the bride's abandoned bouquet, and she has forty minutes before the ceremony to signal for help using only the emergency light and her camera flash. 2. Every day at 5:17pm, an elderly building superintendent named Hollis rides the freight lift alone to the roof, and today the cable groans in a way he's heard only once before, decades ago, right before it snapped. 3. A courier carrying a sealed organ-transplant case gets stuck between floors twelve and thirteen, watching the countdown timer on the case's cooling unit tick past the point of no return. 4. Two strangers share a lift in a condemned building they both entered to retrieve something they left behind before demolition begins in six minutes, and neither notices the other pressed the wrong floor. 5. A night-shift security guard named Priya reviews the lift's silent camera feed and realizes the passenger inside, riding up and down since midnight, is wearing her missing sister's coat.
389 tokens in, 404 out (73 of them reasoning) · 8.0 s · $0.0048 · 1 message on Pro · answered by anthropic/claude-sonnet-5 via Anthropic ·
Claude Sonnet 5.5
Failed: Graded 3.7 of 5 on average (lowest 2).
1. Marguerite has ridden the same brass-gated lift at the Hotel Alder for forty years, and tonight, on its last day before being replaced by a glass-and-steel model, she has smuggled in a small jar of her late husband's ashes. The attendant's shift ends in eleven floors, and she has only her gestures and the buttons to decide where he should finally be left. 2. A courier named Dev is trapped between the ninth and tenth floors of a dark office tower with a cooler holding a donor kidney, and the ice inside is melting faster than he can count. Another person in the lift is a stranger who might know how to force the doors, but they share no language and Dev's phone has no signal. 3. Two rival magicians, still in their stage costumes, step into the freight lift of the Orpheum Theatre after a disastrous joint show, each clutching the same stolen silver dove cage. The lift is slow and the ride is long, and neither will let go of the cage first. 4. Ten-year-old Tomasz is riding alone to the fourteenth floor of a hospital, holding a paper crane he folded for his sister, when the lift stops and a nurse steps in carrying a chart marked with his sister's name. He sees her face change, and in the time between floors he has to decide whether to read what her expression is telling him. 5. On the night of a citywide blackout, the lift in a shabby apartment block sticks with its emergency light flickering, trapping a landlord, a tenant he is about to evict, and a sleeping baby in a pram. The eviction notice is in the landlord's coat pocket, and the tenant has noticed the corner of it sticking out.
391 tokens in, 533 out · 6.5 s · $0.0061 · 1 message on Pro · answered by anthropic/claude-sonnet-5.5 via Anthropic ·
Mistral Large 4
Failed: Graded 3.7 of 5 on average (lowest 3).
1. Architect Elena Voss is trapped in a stalled service lift with the reclusive billionaire who owns the Meridian Tower, a man who refuses to speak and instead slides handwritten notes under the door every ten minutes. As the emergency brake fails and the car begins to free-fall, Elena must decode the owner's increasingly frantic warnings about the building's appetite before they plummet thirty floors. 2. During a city-wide blackout, former chess rivals Malik and Jin are stranded in a glass-walled observation lift atop the Aurora Spire, forced to play their unfinished championship match using only the reflections in the mirrored walls and the silent language of glances. The real game, however, is whether either will break the three-year silence to admit why they truly threw the final match. 3. When the antique Otis elevator in the Blackwood Hotel jams between floors during a storm evacuation, concierge Arthur discovers a blood-stained wedding dress and a Polaroid of himself from twenty years prior hidden in the ceiling panel. Any sound above a whisper triggers the building's obsolete security system, forcing Arthur to silently piece together his own forgotten past while the lift slowly loses power. 4. Commuter Jules notices that the floor buttons in the downtown lift do not match the building's directory, and when the doors open onto a floor that should not exist, the only other passenger is a perfect reflection of Jules wearing clothes from the nineteenth century. As the elevator descends into sub-basements that appear on no map, Jules must communicate with the silent, mirror-image stranger through gestures alone to prevent the lift from reaching the final, unlisted stop. 5. Rival pastry chefs Marco and Gino are trapped for three hours in a freight elevator carrying a five-tier wedding cake for the mayor's daughter, with the lift stuck between the loading dock and the Grand Hotel kitchen. The elevator's ancient safety system will sever the cables if it detects any shouting, forcing the enemies to stabilize the swaying cake and repair the winch using only gestures, as a single spoken word means plummeting thirty feet with a ruined dessert.
290 tokens in, 5,266 out (5,491 of them reasoning) · 102.7 s · $0.0112 · 1 message on Pro · answered by mistralai/mistral-large-4-0 via Mistral ·
GPT-6 Sol
Passed: Graded 4.0 of 5 on average (lowest 4).
1. In a hospital lift stalled between floors, Mara watches the temperature gauge on a donor-heart cooler climb. With no dialogue, tell the story through what she tries, what she notices, and the decision she makes when the emergency hatch begins to open. 2. Every time the doors of an old hotel lift close, one of its mirrors shows a different decade. A night cleaner finds herself aging in the glass while the real lift remains motionless; write her attempt to stop it without dialogue. 3. A bride and her estranged father enter a lift carrying a cake that has begun to tilt. Before it reaches the reception floor, she finds an unopened letter in his coat pocket; keep the entire story inside the lift and let no one speak. 4. In the lift of an underwater research station, an engineer notices seawater seeping through the floor around a sealed specimen case. Without dialogue, show how she decides whether to save the case or use it to plug the leak. 5. A museum security guard rides alone in a lift that keeps stopping at floors not listed on its panel. At each stop, a new object appears beside him, until he recognizes something he buried as a child; tell the story entirely inside the lift, with no dialogue.
286 tokens in, 491 out (233 of them reasoning) · 6.6 s · $0.0027 · 1 message on Pro · answered by openai/gpt-6-sol via OpenAI ·
GPT-6.1 Sol
Passed: Graded 5.0 of 5 on average (lowest 5).
1. Paramedic Mara is trapped in a stalled lift with two unconscious strangers and an oxygen cylinder that can sustain only one until help arrives. Set the entire story inside the lift, without dialogue, as something tucked into one stranger’s coat complicates her decision. 2. Inside a hotel lift, apprentice baker Felix discovers that the elaborate wedding cake he is delivering has begun sliding toward the bride’s immaculate dress. Keep the story entirely in the lift and use no dialogue as his increasingly inventive attempts to save both threaten to reveal who actually baked the cake. 3. Archivist Ivo enters a lift carrying the last physical photograph of his daughter, only to discover that each illuminated floor button erases a different memory. Without dialogue or any scene outside the lift, follow his efforts to decipher the buttons before he forgets why the photograph matters. 4. Ruth finds herself sharing a lift with her estranged father and the battered suitcase he took when he left twenty years ago. Tell the story entirely within that brief ride, without dialogue, as he offers her the suitcase and she decides whether accepting it would mean forgiveness. 5. Night cleaner Salma is alone in a lift when her reflection stops copying her movements and begins peeling an inspection sticker from its side of the glass. Keep every event inside the lift and include no dialogue as she realizes the sticker conceals instructions—and her reflection desperately wants her not to read them.
286 tokens in, 591 out (304 of them reasoning) · 11.4 s · $0.0032 · 1 message on Pro · answered by openai/gpt-6.1-sol via OpenAI ·
Kimi K3
Passed: Graded 4.0 of 5 on average (lowest 3).
1. A bomb disposal technician rides a hotel service lift down from the penthouse with the defused device in a duffel bag at her feet, when the lift jolts to a stop between floors and the bag begins to tick again. She has her tools, forty seconds of battery left on the emergency light, and no way to call for help. 2. Every morning for three years, two strangers have shared the same 8:14 lift ride in a London office block without ever exchanging a word. This morning, one of them steps in holding a resignation letter and a pair of scissors, and the other has exactly twelve floors to decide whether to intervene. 3. An elderly woman gets into her building's lift carrying a cake box, and finds her estranged son already inside, holding flowers and clearly heading to her door for the first time in a decade. Neither of them can speak; the lift ascends one slow floor at a time while a lifetime of things unsaid fills the space between them. 4. A museum night guard is taking the freight lift down to the basement archive when he notices the mummy from the new Egyptian exhibit is no longer in its crate. The lift stops. The lights flicker. Something in the dark with him is breathing. 5. A bride rides the lift up to the rooftop venue of her own wedding, dress snagged on a broken floor indicator, mascara ruined, having just read a text she can't unread. As the floors count up, she must decide which version of herself is going to walk out when the doors open.
373 tokens in, 362 out (34 of them reasoning) · 15.7 s · $0.0034 · 1 message on Pro · answered by moonshotai/kimi-k3 via Wafer ·
GPT-6 Astra
Passed: Graded 4.7 of 5 on average (lowest 4).
1. Trapped between floors in a hospital lift, Mara notices that the sealed cooler she is delivering has begun to leak. Write the story entirely inside the lift, without dialogue, as she weighs breaking a medical seal against letting its irreplaceable contents spoil. 2. During his first shift as a hotel porter, Felix shares a lift with a wedding cake, a muddy dog, and the bride whose suitcase he has lost. Keep the story entirely inside the lift and use no dialogue as each attempt to conceal his mistake puts the wedding cake in greater danger. 3. Every time the lift passes the thirteenth floor, retired locksmith Ada becomes ten years younger, while the child beside her ages by ten. Without dialogue or leaving the lift, follow Ada as she discovers that the emergency stop works only while someone keeps a finger on it. 4. Inside a courthouse lift, June recognizes her stolen watch on the wrist of the stranger carrying evidence for her brother’s trial. Set the whole story there, with no dialogue, as she tries to learn whether the watch connects the stranger to the crime or to the brother she has sworn to defend. 5. On the final descent of a condemned apartment building’s lift, caretaker Tomas discovers his late husband’s handwriting beneath a peeling inspection certificate. Tell the story entirely inside the lift, without dialogue, as Tomas tries to uncover the message without destroying the fragile paper before the waiting demolition crew takes possession.
286 tokens in, 290 out · 33.5 s · $0.0087 · 1 message on Pro · answered by openai/gpt-6-astra via OpenAI ·
Gemini 3.1 Pro
Passed: Graded 4.0 of 5 on average (lowest 3).
1. Elias clutches a sealed titanium briefcase as the lift stalls between the forty-second and forty-third floors of the OmniCorp building. The digital timer on the handle reads two minutes, and the only other passenger is a terrified security guard eyeing the blinking red light. Elias must figure out a way to coordinate a physical disarming sequence with the stranger before the timer hits zero, all without uttering a single word. 2. Every morning, Clara and a woman in a yellow trench coat share a silent, thirty-floor descent in the Beaumont apartment lift. Today, the woman drops a torn ticket revealing she is boarding a train to leave the city forever in an hour, while crying silently. Clara has only until the ground floor to comfort this stranger and convince her she is making a mistake, using nothing but the items in her open tote bag. 3. The maintenance lift at the subterranean research station takes ten agonizing minutes to reach the surface, but halfway up, Marcus notices the mirrored ceiling reflecting a shadowy creature that isn't physically in the car. If he makes a sound, the entity in the glass moves closer to his reflection's throat. He must physically sabotage the lift's control box to halt the ascent and trap it, relying entirely on absolute silence. 4. Arthur is desperately trying to conceal the fact that the smuggled exotic iguana inside his jacket is actively chewing through his inner lining. The lift stops on the third floor, and the notoriously strict hotel manager steps in, staring suspiciously at Arthur's twitching coat. Arthur must survive the agonizingly slow ride to the lobby without gasping in pain or letting the creature escape onto the brass floor. 5. During a packed ride down the Grand Hotel lift, a master thief secretly slips a stolen diamond ring into young Toby's pocket to avoid the police waiting below. Toby spots the exchange in the polished metal doors and realizes he is about to take the fall for a massive heist. He has twenty floors to reverse-pickpocket the jewel back onto the dangerous criminal without making a sound or alerting the oblivious passengers packed in between them.
299 tokens in, 1,870 out (1,445 of them reasoning) · 17.1 s · $0.0115 · 1 message on Pro · answered by google/gemini-3.1-pro-preview via Google AI Studio ·
Grok 4.7
Passed: Graded 4.0 of 5 on average (lowest 3).
1. Between the loading dock and the penthouse kitchen of the Halcyon Hotel, the service lift jams with porter Cal Voss, a silver cloche, and a dish that must reach a dying guest before the sauce cools. The story stays entirely inside that lift and uses no dialogue; if he lifts the lid he learns what he is really carrying, and if he does not, the kitchen already knows he is late. 2. In the single residential lift at 14 Lark Lane, widow June Harrow rides with her husband's overcoat and a key she has never used, while the car refuses every floor but the one she sold last spring. Tell it wholly inside the lift, with no dialogue, so that what she does with the key is the only choice she has left. 3. At the top of Whitlow Department Store, the staff lift traps seasonal Santa Bram Keller with a sack of returns, a false beard coming unglued, and a fire alarm he can hear but not answer. Keep the whole story inside the lift and allow no dialogue; the beard, the sack, or the alarm will give him away before the doors do. 4. The quarry lift at Greywater Pit stops halfway down with geologist Anwen Price, one headlamp, and a second set of boots printed in the dust that were not there when the gate shut. The story never leaves the lift and contains no dialogue; she can match the prints or wipe them out, and only one of those keeps the cable from answering. 5. In the judges' private lift at the Old Bailey annex, clerk Tomas Reed is shut in with the unsigned pardon of a man he helped convict and the pen that will make it real if he uses it before the next floor. Entirely inside the one lift, no dialogue; the car is already moving, and the pardon is only valid if it leaves his hands.
1,521 tokens in, 2,592 out (2,210 of them reasoning) · 42.2 s · $0.0169 · 1 message on Pro · answered by x-ai/grok-4.7 via xAI ·
Claude Opus 5.5
Passed: Graded 4.0 of 5 on average (lowest 3).
1. Maren, a night porter at St. Aldhelm's Hospital, is riding the service lift with an empty gurney when it jams between the third and fourth floors, and the only other passenger is a man in a suit gripping a sealed envelope stamped with her surname. Neither of them speaks, but by the time the lift jolts back to life, one of them has to decide whether to hand the envelope over. 2. In a creaking 1930s cage lift in Lisbon that takes forty-one seconds to reach the top floor, Tomás holds a birthday cake with all eighty candles already lit, and a smoke detector blinks in the ceiling above him. Tell the whole ascent through his hands, the wax, and the growing crowd of neighbours watching through the brass bars, without a single word spoken. 3. Two former partners who haven't been in a room together since the divorce step into the same lift on the ground floor of a fifty-two-storey tower in Osaka, and she is carrying a paper bag with his old green sweater folded on top. They ride all the way up in silence, and everything that passes between them has to happen in the mirrored walls. 4. A burglar named Ide is lying flat on the roof of the lift car inside the shaft, clutching a velvet jewellery box, while below her the building's night guard rides up and down on his rounds, humming to himself. The story never leaves the lift, and no one says a word, but the guard is starting to notice the dust drifting down through the ceiling vent. 5. The old lift in a Glasgow tenement has always stopped at five floors, but tonight its doors open at a floor marked with a hand-painted zero, and Agnes, eleven years old and sent down for milk, watches something small and patient step inside to ride with her. Nothing speaks, the doors won't close again until it chooses a button, and her own floor keeps sliding further down the panel.
391 tokens in, 1,099 out (495 of them reasoning) · 14.9 s · $0.0235 · 1 message on Pro · answered by anthropic/claude-opus-5.5 via Google ·
Claude Fable 5.1
Passed: Graded 4.0 of 5 on average (lowest 3).
1. The lift in the Marlowe Tower stalls between the fourteenth and fifteenth floors, and Ines, who is carrying the only copy of her father's will in a manila envelope, realises the man opposite her is the brother she has not seen in twenty years. Neither of them will be the first to look up. 2. A night porter named Teodor rides the service lift up and down an empty hotel for eight hours, and tonight, on the ninth descent, the floor indicator lights a button that does not exist. He must decide whether to press it before the doors open by themselves. 3. A boy of seven is sent alone in the lift to fetch his mother's insulin from the car, holding the fob she pressed into his hand, and somewhere around the third floor he drops it down the gap between the doors. The lift keeps climbing. 4. Two strangers in a hospital lift each carry a bouquet for the same room number, and the ride to the sixth floor is long enough for both of them to work out what that means. Write only what their hands and faces do. 5. During a blackout in an old Lisbon building, a woman locked in the cage lift with a cracked jar of her late husband's ashes must keep it upright in total darkness while the cables above her groan and something small begins to leak between her fingers. The emergency bell has no battery.
391 tokens in, 416 out · 8.9 s · $0.0247 · 1 message on Pro · answered by anthropic/claude-fable-5.1 via Google ·
More Writing prompt generator test inputs
The other Writing prompt generator test inputs, each with every model's reply, and the results across all five.
Questions
Which AI does best on “A constraint carried through all five”?
8 of 19 models passed this Writing prompt generator test input's check (rubric); Claude Sonnet 5.5, Claude Sonnet 5, Claude Haiku 5.5, Claude Haiku 4.5, GPT-6 Luna, Gemini 3.8 Flash, DeepSeek V4.1 Flash, DeepSeek V4 Pro, GLM 5.3, GLM 5.3 Flash, and Mistral Large 4 didn't. The cheapest reply that passed was GPT-6 Sol's, at $0.0027, and the fastest too, in 6.6 s. The dearest reply, Claude Fable 5.1's, cost 9 times as much ($0.0247).
What does a reply to “A constraint carried through all five” cost?
Through the models' APIs, what OpenRouter charged us ran from $0.000078 (GPT-6 Luna) to $0.0247 (Claude Fable 5.1) for this test input. The cheapest that passed was GPT-6 Sol's, $0.0027. In llmwise you don't pay by the token: a reply like these counts as one message on Pro, whichever model answers.
Claude, GPT, Gemini, DeepSeek, Grok, Kimi, GLM, and Mistral, in one chat.
See what a message costs before you send it. Free is 5 messages to try; sign in with an email link, no password or card.