NC Bench
Overview
About
Tests
Models
Vendors
Run: 2026-02-16-01KHK2J8KTJ317NAC45WTJKAH4
Run Details
Model:
Qwen 3.5 397B A17B
Test:
Dialogue tags
Scenario:
Write 500 words with 70% dialogue
Duration:
3m 18s
Cost:
$0.025843
Provider:
Alibaba
Model input
User/Human
You are a author of comedic heist books. Please write a 500 word scene that has 70% 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 crouched behind the privet hedge, adjusting the strap on her night vision goggles. Bill picked at a clingy burr stuck to his tactical suit, looking miserable in the damp grass. "Tell me again why we are robbing a six-year-old's birthday party," Bill whispered, his voice cracking slightly. "Because the security is lax and the loot density is incredibly high," Molly said, never taking her eyes off the glowing window. "It's Lego, Mol. Pure, unfiltered Lego. Small pieces. Easy to lose. Hard to fence." "Do you have any idea what sealed Lego sets go for on the black market?" "Prison? Maybe a stint in juvie?" "Focus. The target is the pile under the tree. Mostly wrapped. High value items. No cash, but gift cards are liquid." Bill sighed, the sound crunching like dry leaves under his boots. "There are clowns inside. Actual professional clowns. They wear huge shoes." "Clowns are predictable. They twist balloons. They slip on banana peels. They do not carry tasers or guard dogs." "What about the parents? Soccer moms are vicious when their offspring are threatened." "They'll be distracted by the cake cutting ceremony. That is our primary window of opportunity." "And the escape route? Please tell me you have a solid exit strategy." "The bouncy castle in the backyard." Bill stared at her through the darkness. "The bouncy castle. You want us to jump on the inflatable." "It's structural engineering. We slide down the side, grab the sack, and bounce directly into the van." "That sounds like a lawsuit waiting to happen. A civil nightmare." "It's a heuristic diversion. Now, check the perimeter for threats." Bill peeked over the hedge carefully. A balloon animal dog floated past the window slowly. "I see a potential target. Little boy in a cape. He looks vigilant. Watching the door." "Ignore the cape. Watch the gift table. Is the big box still there?" "Yeah. The one wrapped in green dinosaur paper. It's massive." "That's the jackpot. Limited edition robotic T-Rex with sound effects." "You researched the gift registry? Seriously?" "I always research the registry. It is basic heist protocol. Know your loot." "This is insane. We used to hit banks. Vaults. Safety deposit boxes." "Banks have lasers. Kids have juice boxes. Which is actually more dangerous?" "Sugar crashes. Definitely unpredictable sugar crashes." "Exactly. We go in at the pinata smash. Chaos is our cover." "And if we get caught by the birthday boy?" "We claim we are the entertainment. Magicians. Very gritty, street-level magicians." Bill adjusted his glove nervously. "I don't know any magic tricks whatsoever." "Pull a quarter from behind an ear. Improvise. Use sleight of hand." "What if I pull a gum wrapper instead?" "Then you're a trash magician. Just get the bag." "Okay. On the pinata signal. When the candy drops." "Wait. Did you bring the sedatives?" "For the kids? Molly, no." "No, for the parents. Kidding. Just bring the sack." Bill nodded, gripping the burlap tightly in his hands. "I really hope there's no checks. I hate forging signatures." "Cash only. Or gift cards. We fence them at the mall kiosk." "This is the lowest we've sunk, Mol. Truly." "Profit is profit. Ready to move?" Bill looked at the happy house one last time. "As I'll ever be." "Move out."
created_at
"2026-02-16T11:18:15.687874+00:00"
2/16/2026, 11:18:15 AM
model
"qwen/qwen3.5-397b-a17b-20260216"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
893
moderation_latency
(null)
generation_time
191100
tokens_prompt
77
tokens_completion
7380
native_tokens_prompt
76
native_tokens_completion
7166
native_tokens_completion_images
(null)
native_tokens_reasoning
6377
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
0.0258432
router
(null)
provider_responses
0
id
"chatcmpl-fcd26918-90d3-95b5-9b65-3d4c2994bfed"
status
200
is_byok
false
latency
893
endpoint_id
"180484a2-e24b-495a-ba41-b34b680d5a9f"
provider_name
"Alibaba"
model_permaslug
"qwen/qwen3.5-397b-a17b-20260216"
api_type
"completions"
id
"gen-1771240503-pV5qxo17OHqfWXxLu6S3"
upstream_id
"chatcmpl-fcd26918-90d3-95b5-9b65-3d4c2994bfed"
total_cost
0.0258432
cache_discount
(null)
upstream_inference_cost
0
provider_name
"Alibaba"
Evaluation details
Result
Evaluator
Details
Meta Data
0.0000%
Matches word count
n/a
neededClean
false
words
548
0.1879%
Dialogue to Total Word Ratio
Ratio: 81.20%, Deviation: 11.20%
neededClean
false
wordsTotal
548
wordsDialogue
445
0.0940%