At the same time as I was fighting with GPT-5 to give me the thing I actually wanted, I also used the Gemini app and the access I have to pro-level Gemini models to do the Lovecraft test with 2.5-pro. Previously, I had only tested with Flash-2.0, so, not only should this be a more advanced model…well…it should be a more advanced model. I guess that’s it.
Gemini didn’t fight with me. It didn’t suggest edits. It just gave me a story. And the story didn’t revolve around the sea — but it did use the word cyclopean and, I guess, “the sea” is there in the first sentence. But it’s more about a haunted house. I guess. Anyway, the interesting thing here is the approach to the romance — even though it’s a short short story, there’s a shared interest bonding the two main characters, so it seems at least marginally more believable than some of the other models’ attempts. It suffers from the same problem of “how is the narrator narrating this story?” as the GPT-5 story, but it seemed to get the memo much faster…although I don’t know that it sounds very Lovecraftian.
The house stood on a bluff overlooking the cyclopean churn of the sea, its dark windows like vacant eyes staring into the maddening, salt-laced mists. It was there, amidst the decaying splendour of my inherited manor, that I first saw her. Elara, with her raven hair that seemed to absorb the twilight and eyes the color of a starless deep, had come to catalogue the morbid collection of my late uncle—a man whose sanity had long since been sacrificed to the eldritch truths he sought.
Our courtship was not one of sunny days and gentle strolls, but of shared, whispered fascinations. We would spend our nights in the library, poring over my uncle’s forbidden tomes. She was not repulsed, as others had been, by the diagrams of non-Euclidean geometry or the descriptions of entities that swam through the silent gulfs between dimensions. Instead, her delicate fingers would trace the monstrous glyphs with a look of rapt, almost reverent, curiosity. It was in these moments that I felt an unnameable bond, a connection far older and more profound than any terrestrial affection. We were two moths, drawn to the same terrible, consuming flame.
Our love bloomed in the shadow of cosmic dread. One evening, as the moon hung bloated and gibbous over the waves, we found a journal entry hidden behind a false panel. It spoke of a gate beneath the house, a way to a world of endless, silent darkness where time held no meaning. My uncle had never been able to open it, his will having broken before the final rite could be performed. But with Elara by my side, I felt a new strength, a terrifying resolve. Her love, I believed, was the final key, the human sacrifice of mundane sanity that would unlock the profound.
We descended into the cellar, where the air was thick with the scent of damp earth and something far older, something that whispered of un-worlds. The gate was a shimmering, obsidian archway, pulsing with a faint, malevolent light. As I began to chant the words from the journal, Elara’s hand slipped into mine, her grip cool and reassuring. Our eyes met, and in their abyssal depths, I saw not fear, but a shared, ecstatic hunger for the truth.
The gate groaned open, not with a sound, but with a tearing of the very fabric of reality. A cold wind, laden with the dust of dead stars, swept over us, and from the chasm beyond, a voice, a symphony of a thousand maddened choirs, spoke our names. Elara squeezed my hand, her smile a beautiful, terrifying rictus. She wasn’t looking at me, but into the void itself, a look of profound recognition on her face.
And I, in that moment of terrible love, understood. We were never meant to be a couple in the human sense. Our love was not a beginning, but an end. Our union was the final act of a cosmic drama, a sacrifice to the truths that lay waiting beyond the veil. With a shared, blissful scream, we stepped through the gate, our human forms dissolving into motes of star-stuff, our consciousness merging with the cold, all-consuming majesty of the cosmos. We were no longer man and woman, but a singular, infinitesimal thought in the mind of an indifferent, eternal horror. And in that ultimate annihilation, we found a final, dreadful peace.
The readability score agrees with me. It’s less difficult (lower grade level) to read than the GPT-5 version and much easier to read than actual Lovecraft.
( ARLCalc )
AVERAGE READING LEVEL CONSENSUS
The average public reading level in the U.S. is 8th grade. Based on 8 formula(s), your text scored the following:
Score: 9.03 [ = grade level ] Read more
U.S. Grade Level: 9th Grade
Reading Level: Slightly Difficult
Age Range: 14-15
AUTOMATED READABILITY INDEX
Score: 9.93 [ = grade level ]
Reading Difficulty: Somewhat Difficult
Grade Level: 10th Grade
Age Range: 15-16
CALCULATION ON YOUR TEXT
ARI = (4.71 * ( 2640 characters / 568 words)) + (0.5 * (568 words / 30 sentences)) – 21.43
Final result: 9.93
==============================================
FLESCH READING EASE
Score: 66.00 [ = reading scale ] INFO.
Reading Difficulty: Standard
Grade Level: 8th & 9th Grade
Age Range: 13-15 years old
CALCULATION ON YOUR TEXT
206.835 – (1.015 * (568 words / 30 sentences)) – (84.6 * 822 syllables / 568 words)
Final result: 66.00
GUNNING FOG INDEX
Score: 9.20 [ = grade level ]
Reading Difficulty: Slightly Difficult
Grade Level: 9th Grade
Age Range: 14-15
CALCULATION ON YOUR TEXT
FOG = (0.4 * (568 words / (30 sentences + 16 compound sentences)) + 100 * (60 FOG hard words / 568 words))
Final result: 9.20
FLESCH-KINCAID GRADE LEVEL
Score: 8.87 [ = grade level ]
Reading Difficulty: Slightly Difficult
Grade Level: 9th Grade
Age Range: 14-15
CALCULATION ON YOUR TEXT
SCORE = (0.39 * (568 words / 30 sentences)) + (11.8 * (822 syllables / 568 words)) – 15.59
Final result: 8.87
COLEMAN-LIAU READABILITY INDEX
Score: 8.90 [ = grade level ]
Reading Difficulty: Slightly Difficult
Grade Level: 9th Grade
Age Range: 14-15
CALCULATION ON YOUR TEXT
SCORE = (0.0588 * (2537 letters / 568 words) * 100) – (0.296 * (30 sentences / 568 words) * 100) – 15.8
Final result: 8.90
==============================================
THE SMOG INDEX
Score: 8.29 [ = grade level ]
Reading Difficulty: Average – Slightly Difficult
Grade Level: 8th Grade
Age Range: 13-14
CALCULATION ON YOUR TEXT
SCORE = 1.043 * Sqrt((60 hard words * (30 / 30 sentences)) + 3.1291)
Final result: 8.29
ORIGINAL LINSEAR WRITE FORMULA
Score: 76.00 [ = reading scale ] INFO.
Reading Difficulty: Slightly Difficult
Grade Level: 9th Grade
Age Range: 14-15 years old
CALCULATION ON YOUR TEXT
Original Version Score = ((395 one-syllable words – 55 ignored words) * 1) + (3 * (30 sentences + 0 compound sentences)) / (568 total words / 100))
Final result: 75.70 = 76
==============================================
LINSEAR WRITE GRADE LEVEL FORMULA
Score: 10.55 [ = grade level ]
Reading Difficulty: Fairly Difficult
Grade Level: 11th Grade
Age Range: 16-17
CALCULATION ON YOUR TEXT
Initial Score = (((508 easy words – 55 ignored words) * 1) + (60 hard words * 3)) / (30 sentences + 0 compound sentences)
Adjusted Score = 21.10 (initial score) / 2
Final result: 10.55
==============================================
FORCAST READABILITY FORMULA
Score: 9.57 [ = grade level ]
Reading Difficulty: Somewhat Difficult
Grade Level: 10th Grade
Age Range: 15-16
CALCULATION ON YOUR TEXT
SCORE = 20 – ((395 1-syllable words * 150) / (568 words * 10))
Final result: 9.57
Results calculated on 8/10/2025 at 10:14:42 PM by ReadabilityFormulas.com
Also, I felt like Gemini (who produced the image above) didn’t do nearly as good a job with rendering an image for the story as GPT does. It suffers from a lot of the WTFs that are common with AI-generated art — particularly architecture that doesn’t make sense, but also, I think it’s notable that, again, we’ve returned to having the two subjects in silhouette — a thing I didn’t even notice was a thing AI did until I saw the image GPT-5 produced for the last story.
I’ve been using Gemini in VS Code and it seems to do…okay. I can’t say that it’s significantly better or worse than ChatGPT. And, in the context of this test, I feel largely the same. It’s pretty meh. But, I was able to produce some unintentionally hilarious videos when I tested out Veo…
Leave a Reply