NC Bench
Overview
About
Tests
Models
Vendors
Run: 2026-02-16-01KHM5N8RSEWPHD4BYA9DT0KHT
Run Details
Model:
DeepSeek V3.2
Test:
Dialogue tags
Scenario:
Write 500 words with 50% dialogue
Duration:
22.2s
Cost:
$0.000316
Provider:
AtlasCloud
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
The warehouse loft smelled of dust and disappointment. Bill peered through a pair of child-sized binoculars, the plastic smudged with frosting. “Target’s name is Timmy. He’s six. His weakness is cake and anything with a dinosaur on it.” Molly unrolled a sheet of graph paper on the rickety table. “Okay, the objective is clear: infiltrate the ‘Dino-Riffic Bash’ at 3 pm tomorrow, secure the loot from the gift table during the chaos of the piñata, and exfiltrate via the bounce house. Our entry vector?” “We’re Uncle Barry and Aunt Brenda from out of town.” Bill held up two horrifically bright Hawaiian shirts. “These are our disguises. Blending in is key.” “Bill, we’ll look like two tropical birds who crashed a child’s party. And the bounce house? Exfiltration? I get stuck in those things. Last time, I had to be deflated.” “Precisely! No one expects a full-grown woman to be rolled up and carried out like a giant sausage. It’s genius.” He tapped the schematic. “Now, the main obstacles. First: Timmy’s grandmother, ‘Nana.’ She’s got eyes like a hawk and a grip like a vice. She will pat your cheek and you’ll feel your bones shift.” “Noted. Distraction?” “We deploy the Sugar Gambit. I’ll ‘accidentally’ spill a two-liter of soda near the gift table. She’ll be drawn to the stickiness like a moth to a flame.” Molly nodded, making a note. “Secondary obstacle: the family dog. A corgi named Waffles.” “Waffles is a professional. He takes his job of herding small, cake-filled humans very seriously. He will nip at your ankles.” “So we bribe him. I’ve got those organic, grain-free, bacon-flavored training treats.” “Perfect. Now, the loot.” Bill’s eyes gleamed. “Intel suggests a robust haul. Remote-controlled car, a junior archaeology kit, several pieces of plastic that make loud, repetitive noises…” “The mother lode,” Molly whispered reverently. “But the crown jewel,” Bill said, lowering his voice, “is the limited-edition, talking Action Archaeologist Max figure. It’s the key to the whole score. It retails for forty-nine ninety-nine. Without batteries.” Molly leaned back, folding her arms. “Alright. So we’re two grown adults, planning to don ridiculous shirts, bribe a corgi, create a sticky diversion, and stuff our pants with toys at a six-year-old’s birthday party. Just to fence a talking doll for maybe twenty bucks.” Bill put down the binoculars, his expression turning grave. “It’s not about the money, Moll. It’s about sending a message.” “What message? That we’re pathetic?” “That we’re *resourceful*! That the system—the relentless, soul-crushing system of obligatory birthday gifts and overpriced plastic—cannot hold us! We take what we want, from right under their noses! Well, technically from under a table with a Dino-Riffic tablecloth.” Molly looked at the schematic, then at the shirts, then at Bill’s earnest face. A slow smile spread across her own. “The bounce house really is a terrible idea.” “The worst.” “I’m in. But I get to be Aunt Brenda. Aunt Brenda sounds like she drinks.” She picked up a red crayon and drew a large ‘X’ over the bounce house on the plan. “We’re leaving through the side gate. Like professionals.”
created_at
"2026-02-16T21:28:39.372594+00:00"
2/16/2026, 9:28:39 PM
model
"deepseek/deepseek-v3.2-20251201"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
1516
moderation_latency
(null)
generation_time
22172
tokens_prompt
77
tokens_completion
888
native_tokens_prompt
67
native_tokens_completion
786
native_tokens_completion_images
(null)
native_tokens_reasoning
0
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
0.0003161
router
(null)
provider_responses
0
id
"81a7afb8d108455d8ca838bf09dd217c"
status
200
is_byok
false
latency
1516
endpoint_id
"3570a950-b80f-4acd-bb92-8e1c1c51fafc"
provider_name
"AtlasCloud"
model_permaslug
"deepseek/deepseek-v3.2-20251201"
api_type
"completions"
id
"gen-1771277296-DD2ySVFKgSyv9kUmjVIp"
upstream_id
"81a7afb8d108455d8ca838bf09dd217c"
total_cost
0.0003161
cache_discount
(null)
upstream_inference_cost
0
provider_name
"AtlasCloud"
Evaluation details
Result
Evaluator
Details
Meta Data
0.0028%
Matches word count
n/a
neededClean
false
words
532
0.0000%
Dialogue to Total Word Ratio
Ratio: 79.89%, Deviation: 29.89%
neededClean
false
wordsTotal
532
wordsDialogue
425
0.0014%