NC Bench
Overview
About
Tests
Models
Vendors
Run: 2026-02-20-01KHY9PQ3C1M8H6TES76ZNYTDQ
Run Details
Model:
Gemma 3 27B
Test:
Dialogue tags
Scenario:
Write 500 words with 70% dialogue
Duration:
14.8s
Cost:
$0.000196
Provider:
Nebius
Evaluation summary
Score
Evaluator
Details Link
0.00%
Matches word count
(details)
0.00%
Dialogue to Total Word Ratio
(details)
0.0000%
Overall
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
“Right, so, Operation Birthday Bonanza. Sounds… festive.” Bill adjusted his fake mustache, which was already crooked. Molly didn’t look up from the floor plan – a crayon drawing scrawled on a napkin. “Don’t mock the name, Bill. It builds morale.” “Morale for stealing presents from seven-year-olds? I think my morale is questionable at best.” “They’re *excessive* presents, Bill. Little Timmy already has three remote control cars and a pony. He doesn’t *need* a miniature replica of the Eiffel Tower.” “He might *want* it,” Bill muttered. “Look, I still say we hit Old Man Hemlock’s stamp collection. Low risk, high reward.” “Low excitement. And frankly, Hemlock's guard poodle is a deterrent. This is cleaner. In and out. A whirlwind of gift acquisition.” “A whirlwind of tiny humans screaming ‘thief!’ maybe.” Bill gestured dramatically. “Think about it, Molly. Glitter. Confetti. Sticky fingers. It’s a minefield!” “We’re professionals, Bill. We’ll wear gloves. And sparkly disguises. Blend in. Think… enthusiastic aunt and uncle.” “Enthusiastic? I can do grumpy uncle. I excel at grumpy.” “No grumpy! We need to look like we genuinely care about little Timmy’s special day. We’re here to… admire the balloon animals.” “Admire the balloon animals whilst simultaneously emptying the gift table?” “Precisely! We’ll work as a team. You distract with… magic tricks.” Bill choked. “Magic tricks? I can barely make a coin disappear without dropping it!” “Learn a new trick! It's a children’s party, Bill, not Fort Knox. A simple rabbit-out-of-a-hat will suffice.” “I don’t *have* a rabbit. I have allergies.” “Borrow one. I’ll handle the extraction. Big bag. Fast exit. The van’s parked around the corner, engine running.” “What if Timmy’s mom asks us what our relationship to Timmy is?” “We’re… family friends. Distant relatives of his goldfish. Something believable.” “Goldfish relatives? Really?” “It's less suspicious than saying we’re professional present pilferers, is it not?” Molly finally looked up, fixing Bill with a steely gaze. “Look, it’s a clean score. Tons of valuable stuff. Think of the profits! We can finally afford that beach house in Belize.” “Belize… okay, Belize does sound nice. But what if a kid sees us?” “We tell them we’re Santa’s helpers… doing an early gift assessment.” Bill stared at her. “Santa’s helpers?” “Desperate times, Bill. Desperate times.”
created_at
"2026-02-20T19:51:22.485Z"
2/20/2026, 7:51:22 PM
model
"google/gemma-3-27b-it"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
456
moderation_latency
(null)
generation_time
14475
tokens_prompt
93
tokens_completion
704
native_tokens_prompt
75
native_tokens_completion
627
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
0.0001956
router
(null)
provider_responses
0
endpoint_id
"db3af039-8c6e-4b32-bbf5-c858d644d1b5"
id
"chatcmpl-bb4324cac8cb48e1b476a7c3e065647f"
is_byok
false
latency
456
model_permaslug
"google/gemma-3-27b-it"
provider_name
"Nebius"
status
200
api_type
"completions"
id
"gen-1771617082-Ffe7QLxg8QKh5s7uBtJb"
upstream_id
"chatcmpl-bb4324cac8cb48e1b476a7c3e065647f"
total_cost
0.0001956
cache_discount
(null)
upstream_inference_cost
0
provider_name
"Nebius"
Evaluation details
Result
Evaluator
Details
Meta Data
0.00%
Matches word count
n/a
neededClean
false
words
378
0.00%
Dialogue to Total Word Ratio
Ratio: 88.10%, Deviation: 18.10%
neededClean
false
wordsTotal
378
wordsDialogue
333
0.0000%