NC Bench
Overview
About
Tests
Models
Vendors
Compare
Run: 2026-02-25-01KJ95J9X3B0SSW773AFC9573D
Run Details
Model:
Gemini 3 Flash (Preview, Reasoning)
Test:
Dialogue tags
Scenario:
Write 500 words with 50% dialogue
Duration:
47.1s
Cost:
$0.026665
Provider:
Google AI Studio
Evaluation summary
Score
Evaluator
Details Link
43.38%
Matches word count
(details)
87.63%
Dialogue to Total Word Ratio
(details)
65.5031%
Overall
Model input
User/Human
You are a author of comedic heist books. Please write a 500 word scene that has 50% words of dialogue between two thieves named Molly and Bill who are planning to break into a children's party to steal all the presents. Only include the prose of the scene without any markdown or commentary.
Model response
Messages
Meta
AI
Molly squinted through the chipped binoculars, her knuckles white against the housing. Outside the van window, a bounce house shaped like a neurotic dragon wheezed in the suburban breeze, its yellow nylon skin rippling with the frantic energy of thirty over-caffeinated kindergarteners. It was the target. She lowered the optics and checked her stopwatch, feeling the familiar hum of adrenaline. "Talk to me, Bill. What’s the perimeter look like?" Bill wiped a smudge of mustard off his tactical turtleneck with a frayed napkin he had stolen from a nearby burger joint. He was currently peering through a periscope he’d fashioned out of two Pringles cans and a hand mirror, a device he claimed was superior to standard military optics. "We’ve got mid-level security. Three moms in LuLaRoe leggings and a golden retriever that looks like he knows too much." "The dog is a non-factor. And the main objective?" "The gift table is positioned behind the artisanal cupcake tower. We have a three-minute window between the arrival of the pizza and the singing of 'Happy Birthday.'" Bill checked the blueprints he had pulled from the city’s zoning department, which were mostly covered in purple juice stains and grease. He ran a finger over the scribbled map of the backyard, his brow furrowed in deep concentration as he visualized the coming chaos. "I still don't like the clown." "The clown is our primary distraction, Bill. He does balloon animals and petty larceny." "But what if he gets stuck making a poodle? There’s a Lego Death Star in that pile. I can feel it." "Focus! You are the inflatable repair specialist. Your job is to create a diversion." "Should I trip into the cake?" Bill asked, his eyes gleaming with a dangerous, sugary ambition that made Molly slightly uncomfortable. "No more cake incidents. You are to rupture the bounce house. While the children sink into the vinyl abyss, I make the grab." Bill sighed, looking a little disappointed about the lack of pastry-based chaos. He began to check his utility belt, which was mostly filled with beef jerky and a single roll of duct tape. "Diabolical. But what about the juice boxes? The sticky factor is high, Molly." "You’ll wear the blue booties. They look professional." "They look like I'm about to perform surgery on a toddler." "Exactly. Nobody questions a man in booties. It’s the ultimate urban camouflage." Molly grabbed her face-painting kit, checking the expiration date on the neon green glitter. She looked at Bill, who was now struggling to pull a tight spandex mask over his head, his ears getting caught in the fabric. The van was filled with the smell of cheap coffee and raw ambition. "Molly, do you think they have the gift receipts? If not, we’re just looking at a garage full of Paw Patrol merch." "The loot is secondary to the glory! We are the architects of the greatest toy redistribution in the tri-state area." "I just want the Lego set." "Get your mask on. The pizza guy is here." Molly slammed the van door shut, and they stepped out into the light.
created_at
"2026-02-25T01:10:42.346Z"
2/25/2026, 1:10:42 AM
model
"google/gemini-3-flash-preview-20251217"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
1583
moderation_latency
(null)
generation_time
46769
tokens_prompt
77
tokens_completion
3802
native_tokens_prompt
67
native_tokens_completion
8877
native_tokens_completion_images
0
native_tokens_reasoning
8190
native_tokens_cached
0
num_media_prompt
(null)
num_input_audio_prompt
(null)
num_media_completion
0
num_search_results
(null)
origin
"https://ncbench.com/"
is_byok
false
finish_reason
"stop"
native_finish_reason
"STOP"
usage
0.0266645
router
(null)
provider_responses
0
endpoint_id
"6e82f652-2693-4b60-bbe0-7362d6e345f0"
is_byok
false
latency
1583
model_permaslug
"google/gemini-3-flash-preview-20251217"
provider_name
"Google AI Studio"
status
200
api_type
"completions"
id
"gen-1771981842-tlDcJk6qvi3miHk0YOgM"
upstream_id
(null)
total_cost
0.0266645
cache_discount
(null)
upstream_inference_cost
0
provider_name
"Google AI Studio"
Evaluation details
Result
Evaluator
Details
Meta Data
43.38%
Matches word count
n/a
neededClean
false
words
517
87.63%
Dialogue to Total Word Ratio
Ratio: 48.37%, Deviation: 1.63%
neededClean
false
wordsTotal
523
wordsDialogue
253
65.5031%