NC Bench
Overview
About
Tests
Models
Vendors
Run: 2026-02-16-01KHM63DX726CCHZ0BNF56SH8V
Run Details
Model:
DeepSeek V3 (2025-03-24)
Test:
Dialogue tags
Scenario:
Write 500 words with 30% dialogue
Duration:
4.0s
Cost:
$0.001521
Provider:
Nebius
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
The warehouse was drafty, the kind of place where even the rats wore tiny sweaters. Molly tapped her foot against a crate labeled *Fragile: Probably Just More Cardboard*, while Bill squinted at a hand-drawn map of the party venue—a sprawling mansion owned by the city’s most insufferable toy magnate, Reginald P. Twinkle. "You sure this is the right place?" Molly asked, flicking a cobweb off her sleeve. Bill grinned. "Positive. Twinkle’s annual ‘Kiddie Extravaganza’ is tomorrow. Every spoiled brat in town will be there, and so will their presents." Molly sighed. "Breaking into a children’s party feels… low." "Low? Molly, this is *artistry*. We’re not stealing from kids—we’re stealing from *parents* who buy their kids too much junk." She rolled her eyes. "Fine. What’s the plan?" Bill spread the map wide, revealing a maze of ball pits, bouncy castles, and a suspiciously large area labeled *Cotton Candy Cannon Zone*. "We go in disguised as clowns—" "Clowns?" Molly groaned. "—because no one looks at clowns twice. Too scary. We blend in, load the presents into a piñata-shaped getaway vehicle, and vanish before anyone notices." Molly stared. "You built a piñata-shaped car?" "It’s a *truck*. And yes." She pinched the bridge of her nose. "What’s the catch?" Bill hesitated. "Well… Twinkle hired a security detail. Former circus performers." "Of course he did." "But!" Bill held up a finger. "They’re allergic to confetti. One good blast from a confetti cannon, and they’ll be too busy sneezing to stop us." Molly considered this. "You’re an idiot." "A *brilliant* idiot." She sighed again, deeper this time. "Fine. But if we get caught, I’m telling everyone this was your idea." Bill clapped his hands. "Deal! Now, let’s practice our clown laughs." Molly deadpanned. "No." "Come on! A good heist needs chemistry!" She glared. "The only chemistry here is the kind that’ll explode in our faces." Bill ignored her, launching into a wheezing, over-the-top cackle that echoed through the warehouse. Somewhere, a pigeon took off in panic. Molly massaged her temples. "We’re doomed." "Nonsense!" Bill slapped her on the back. "By this time tomorrow, we’ll be rolling in loot. And by loot, I mean action figures and those weird jelly shoes kids love." Molly muttered, "I hate my life." Bill grinned. "No, you hate *boring* heists. This? This is legendary." She didn’t argue. Mostly because she knew he was right.
created_at
"2026-02-16T21:36:05.428262+00:00"
2/16/2026, 9:36:05 PM
model
"deepseek/deepseek-chat-v3-0324"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
708
moderation_latency
(null)
generation_time
4002
tokens_prompt
77
tokens_completion
685
native_tokens_prompt
66
native_tokens_completion
654
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
0.001521
router
(null)
provider_responses
0
id
"chatcmpl-0e74d963bb6844819c1a5fec48eadd40"
status
200
is_byok
false
latency
708
endpoint_id
"0dd6bc5a-8695-4d4d-8067-2f4ccf7e49e0"
provider_name
"Nebius"
model_permaslug
"deepseek/deepseek-chat-v3-0324"
api_type
"completions"
id
"gen-1771277760-k1qJMW4AYBdYQFIebhT0"
upstream_id
"chatcmpl-0e74d963bb6844819c1a5fec48eadd40"
total_cost
0.001521
cache_discount
(null)
upstream_inference_cost
0
provider_name
"Nebius"
Evaluation details
Result
Evaluator
Details
Meta Data
0.0000%
Matches word count
n/a
neededClean
false
words
401
0.0000%
Dialogue to Total Word Ratio
Ratio: 49.88%, Deviation: 19.88%
neededClean
false
wordsTotal
401
wordsDialogue
200
0.0000%