NC Bench
Overview
About
Tests
Models
Vendors
Run: 2026-02-20-01KHY9KH57S4G10F4W53WKM8HB
Run Details
Model:
Gemma 3 27B
Test:
Dialogue tags
Scenario:
Write 500 words with 70% dialogue
Duration:
24.4s
Cost:
$0.000105
Provider:
DeepInfra
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 let me get this straight,” Bill said, meticulously polishing his spectacles with a napkin pilfered from a diner three towns over. “We’re… robbing a birthday party.” Molly rolled her eyes, already dressed in a suspiciously bright floral jumpsuit. “It’s not ‘robbing,’ Bill. It’s… *reallocating* gifts. And it’s a very lucrative opportunity. Little Timmy’s turning six. Six! That’s a significant gift haul.” “Lucrative because people buy small children expensive things they don't appreciate? That feels… wrong, even for us.” “Wrong is letting perfectly good LEGO sets gather dust when we could be, you know, funding our early retirement,” Molly retorted. “Think of the beaches, Bill. The tiny umbrellas in the drinks!” “I'm picturing tiny handcuffs,” Bill muttered. “And a very angry group of parents.” “Relax. We’re professionals. Mostly. It’s a simple in-and-out. I've scouted the location. It’s at ‘Bouncy Castle Bonanza,’ which, frankly, is a security nightmare. Too much focus on inflatable structures, not enough on… present security.” “They have a bouncy castle shaped like a dinosaur,” Bill stated, consulting a hand-drawn map riddled with crayon markings. “Irrelevant! The presents are in a back room, guarded by… a teenager in a wizard hat.” “A wizard hat?” “Apparently, it’s part of the ‘enchantment’ theme. He’ll be too busy attempting card tricks to notice our sophisticated operation.” “Sophisticated? You're planning on distracting him with a glitter bomb, Molly. That's hardly sophisticated.” “It’s highly effective! And sparkly. Besides, you're on distraction duty. You're going to challenge him to a magic duel.” Bill choked on air. “Me? A magic duel? I can barely make a coin disappear without dropping it.” “Details, details. Fake confidence, Bill. You can do it. Tell him you studied under Merlin himself.” “I studied under a tax accountant, Molly.” “Same difference! Look, the time is 14:00 hours. Peak party chaos. Kids sugared up on cake, parents distracted by small human tornadoes. We’re in, we grab the loot, we’re out. Remember the bag? The oversized Santa sack?” “It’s a bit… conspicuous, isn’t it?” “It’s thematic! It blends in! It screams ‘harmless grandparent!’ Plus, it holds a lot of presents. We’re talking potential for a new yacht here, Bill.” “I still prefer the beaches and tiny umbrellas.” “We can have both! Now, practice your dramatic wizard-dueling pose. And remember, no tripping over bouncy castles.”
created_at
"2026-02-20T19:49:38.097Z"
2/20/2026, 7:49:38 PM
model
"google/gemma-3-27b-it"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
459
moderation_latency
(null)
generation_time
24276
tokens_prompt
93
tokens_completion
710
native_tokens_prompt
75
native_tokens_completion
617
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
0.00010472
router
(null)
provider_responses
0
endpoint_id
"8f22002c-c045-446f-a1b9-9896133536b8"
id
"chatcmpl-R0Vg2fQsVE3cWvINQ5FT13Jg"
is_byok
false
latency
49
model_permaslug
"google/gemma-3-27b-it"
provider_name
"DeepInfra"
status
200
api_type
"completions"
id
"gen-1771616978-HRflzKgjlbOhwQRXvjxj"
upstream_id
"chatcmpl-R0Vg2fQsVE3cWvINQ5FT13Jg"
total_cost
0.00010472
cache_discount
(null)
upstream_inference_cost
0
provider_name
"DeepInfra"
Evaluation details
Result
Evaluator
Details
Meta Data
0.00%
Matches word count
n/a
neededClean
false
words
387
0.00%
Dialogue to Total Word Ratio
Ratio: 87.86%, Deviation: 17.86%
neededClean
false
wordsTotal
387
wordsDialogue
340
0.0000%