NC Bench
Overview
About
Tests
Models
Model
Z.AI GLM 4.6
Test
Dialogue tags
Scenario
Write 500 words with 30% dialogue
Run ID
2026-02-08-01KGZ0BDW3NR04ATWDTFVRZHSZ
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 blueprints were a work of art, spread across the greasy hood of the van. Molly, her face illuminated by a single battery-powered lamp, traced a gloved finger over the layout of a suburban backyard. Across the street, a brightly coloured bouncy castle sagged like a defeated dragon, its air pump humming a mournful tune. Bill, on the other hand, was meticulously arranging a collection of cheese puffs into a tiny, orange pyramid. “The target is a ‘Timmy’s 7th Birthday Extravaganza’,” Molly hissed, pointing at the schematic. “The parents are distracted by the bouncy castle. That’s our window.” “Do you think they’ll have those little hot dogs? The ones in the barbecue sauce?” Bill asked, wiping a smear of orange dust from his chin strap. “We are not there for the canapés, Bill. We are there for the payload. The presents.” “Right, right. The loot. I call dibs on anything with a ninja on it.” “You’ll get what you’re given. Now, the primary objective is the gift table, located adjacent to the unicorn-shaped cake. I’ll deploy the Acoustic Distraction Device.” “The party horn? Cool.” “It’s a sophisticated tool of misdirection. While they’re looking for the phantom honker, you use the grappling hook to snag the goodie bags from the tree.” Bill’s eyes lit up. “What about the piñata? That thing’s gotta be loaded.” “It’s a high-risk, low-reward scenario, Bill. Too much shrapnel. We stick to the plan. In and out in ninety seconds, before they even cut the cake.” Molly tapped a small, silver canister. “I’ll use the Cryo-Lock Disengager on the back gate if we need a quick exit.” It was a can of compressed air for cleaning keyboards, but the label had been peeled off. Bill nodded sagely, impressed. He picked up a grappling hook, which looked suspiciously like a giant, metal coat hanger tied to a rope. He gave it an experimental swing, nearly taking out the van’s side mirror. Molly sighed, a sound of profound and eternal suffering. This was the price of working with a professional.
created_at
"2026-02-08T16:12:51.834574+00:00"
Sunday, February 8th, 2026 at 4:12:51 PM GMT+00:00
model
"z-ai/glm-4.6"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
239
moderation_latency
(null)
generation_time
83063
tokens_prompt
61
tokens_completion
2327
native_tokens_prompt
66
native_tokens_completion
2342
native_tokens_completion_images
(null)
native_tokens_reasoning
1910
native_tokens_cached
32
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
0.0040922599999776
router
(null)
provider_responses
0
id
"chatcmpl-RZ2RopsnJd1QMI7lhRoobS5U"
status
200
is_byok
false
latency
65
endpoint_id
"c8607edc-8577-4199-b03d-1213da4743dc"
provider_name
"DeepInfra"
model_permaslug
"z-ai/glm-4.6"
api_type
"completions"
id
"gen-1770567088-oyfTD7KWkNF436MTTZt2"
upstream_id
"chatcmpl-RZ2RopsnJd1QMI7lhRoobS5U"
total_cost
0.0040922599999776
cache_discount
0.0000112000000224
upstream_inference_cost
0
provider_name
"DeepInfra"
Evaluation details
Result
Evaluator
Details
Meta Data
0.0000%
Matches word count
n/a
neededClean
false
words
346
0.0000%
Dialogue to Total Word Ratio
Ratio: 50.00%, Deviation: 20.00%
neededClean
false
wordsTotal
346
wordsDialogue
173
0.0000%