NC Bench
Overview
About
Tests
Categories
Models
Vendors
Compare
Run: 2026-03-25-01KMJ8FAPCY1CT2BYNSGADWKGV
Run Details
Model:
Qwen 3 32B
Test:
Dialogue tags
Scenario:
Write 500 words with 30% dialogue
Duration:
17.1s
Cost:
$0.000624
Provider:
Novita
Evaluation summary
Score
Evaluator
Details Link
0.21%
Matches word count
(details)
0.00%
Dialogue to Total Word Ratio
(details)
0.1070%
Overall
Model input
User/Human
You are a author of comedic heist books. Please write a 500 word scene that has 30% 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 adjusted her giant foam sword, frowning at the bounce house deflating outside the warehouse where Bill waited in a unicorn onesie. “You’re sure this is a *party* and not a toddler’s dress-up prison?” he asked, tugging at the frilly saddle. She kicked a neon My Little Pony costume at her feet. “It’s a ‘princess-themed’ bash. The parents paid extra for the glitter. Our in is the piñata.” Bill gaped. “A piñata? Molly, those are paper mache and *sugar*. How does that help us steal fifty Carebears?” She handed him a mallet. “We *break* the piñata. Distract the kids with the candy rain while I disable the alarm—wired to the bounce house’s plug, per the blueprints.” She gestured to her hair, now secured under a tiara that squeaked when she moved. “And you’ll be our chaos cannon.” “And by chaos cannon you mean—?” “Throwing these.” She shoved a handful of whoopee cushions at him. “Stick one on every chair. When Mommy Dearest goes to investigate the ‘mysterious noise,’ that’s our window to hit the gift table.” Bill slumped. “What if we get tackled by a five-year-old?” “Then you’ll yell ‘UNICORN应急计划!’ and sprint for the exit.” “‘Unicorn emergency plan’ is in *Chinese*?” “Mandarin. You practice. While I’m busy juggling a toddler’s juice box and the alarm code, you’d better not freeze like a baked potato.” The warehouse door creaked open. Through the window, a pink banner read *PRINCESS SOPHIE’S 6TH BIRTHDAY!* A crowd of kids clamored over a castle-shaped cake table. Molly crouched behind a cardboard castle, her sword clanking. Bill fumbled with his mallet. “Remember,” she whispered, “the plan is simple: hit the piñata, deploy the whoopees, cause confusion, and vanish before someone realizes all the presents are missing.” Bill salivated. “Vanish how? I can’t fit in that tiara without looking like a rejected Disney villain.” She heaved the bounce house’s plug toward him. “Tripwire it to this. When the party chaos peaks, pull the cord. The bounce house deflates, we dive inside, and roll out the back like a gifted balloon animal.” He stared at the plug. “Molly, that thing’s rigged to blow *up* if you yank it wrong.” “Not if you yank it *right*. Now—” She nodded toward the window where a kid had just climbed onto the piñata, rope in hand. “Your cue.” Bill sprinted forward, mallet raised, as Molly inched toward the gift table. The piñata swung; the kid missed. Bill whacked it harder than intended. A shower of confetti and gummy worms rained down. Kids shrieked. “WHOOOPIE CUSHIONS, ACTIVATE!” Bill barked, slapping them onto chairs. A mother gasped as she sat down. Molly was at the gift table, untying a bow, when Bill barreled back, covered in cake. “The bounce house—it’s inflating *more*?!” She snatched a bag of stolen robots. “Pull the plug *now*.” He yanked it. The bounce house lurched upward, trapping them inside. They tumbled out the back, laughing, as the warehouse erupted in screams of “UNICORN!” and “SABOTAGE!” “Next job,” Bill wheezed, “let’s hit a *dinosaur* party. I know how those work.” Molly tossed him a stolen kazoo. “Dinosaurs are too loud. We need something with *more* glitter.”
created_at
"2026-03-25T10:26:04.113Z"
3/25/2026, 10:26:04 AM
model
"qwen/qwen3-32b-04-28"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
231
moderation_latency
(null)
generation_time
16982
tokens_prompt
93
tokens_completion
1523
native_tokens_prompt
73
native_tokens_completion
1370
native_tokens_completion_images
(null)
native_tokens_reasoning
622
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
0.0006238
router
(null)
provider_responses
0
endpoint_id
"d342f1c5-34b5-4a5c-b2ea-a0cdcbf72e70"
id
"75245fb5-80a4-46b6-a569-7f4d8ba76ad6"
is_byok
false
latency
231
model_permaslug
"qwen/qwen3-32b-04-28"
provider_name
"Novita"
status
200
user_agent
"langchainjs-openai/1.0.0 ((node/v24.14.0; linux; x64))"
http_referer
(null)
api_type
"completions"
id
"gen-1774434364-IA0Nj7jCZs2ss0e2ED6O"
upstream_id
"75245fb5-80a4-46b6-a569-7f4d8ba76ad6"
total_cost
0.0006238
cache_discount
(null)
upstream_inference_cost
0
provider_name
"Novita"
Evaluation details
Result
Evaluator
Details
Meta Data
0.21%
Matches word count
n/a
neededClean
false
words
528
0.00%
Dialogue to Total Word Ratio
Ratio: 52.63%, Deviation: 22.63%
neededClean
false
wordsTotal
532
wordsDialogue
280
0.1070%