NC Bench
Overview
About
Tests
Categories
Models
Vendors
Compare
Run: 2026-09-02-01M1GQJK53PV0279ADZP6FVVWE
Run Details
Model:
Z.AI GLM 5.3 Flash (Reasoning, Max)
Test:
Dialogue tags
Scenario:
Write 500 words with 30% dialogue
Duration:
4m 47s
Cost:
$0.003109
Provider:
Z.AI
Evaluation summary
Score
Evaluator
Details Link
100.00%
Matches word count
(details)
95.31%
Dialogue to Total Word Ratio
(details)
97.6534%
Overall
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 van smelled like stale cheese crackers and poor decisions. Bill had his laptop propped on a milk crate, running a slideshow he had clearly spent all week on. There were transitions. There was a title slide: OPERATION CANDLE BLOWOUT, in a font that spun. "Target," he announced, clicking to a long-lens photograph of a suburban backyard, "is the Kowalski party. Saturday. Thirty-two presents." Molly stared at him for a long moment. "We're casing a six-year-old." "Seven. She's advanced." The next slide showed the gift table: plastic dinosaur tablecloth, a promising mountain of boxes. Bill's red laser dot trembled across it with genuine feeling. Then came a laminated slide labeled THREAT MATRIX, which ranked the twenty-two invited children by tattletale potential. A girl named Destiny was flagged in red. Under NOTES, Bill had written: never turns her back. Below the children, in cheerful clip-art, was a golden retriever named Biscuit, whose dossier listed exactly one weakness. "With what?" Molly asked, when Bill got there. "Hot dogs. I budgeted for hot dogs." He said it like a man who had personally bribed a warden. Then a hand-drawn diagram of the yard, annotated in three colors of pen. Chain-link fence, hedge with a gap, and one enormous circle in the middle that Molly hoped was a trampoline. "What's the circle?" "Bouncy castle." "Ah." "The whole operation runs through it. Motor's loud. Covers the fence cut." He clicked to a timeline. "Kids hit cake at one-fifteen. We get four minutes. The piñata is our hard out—once that thing swings, eyes go up, and we're gone." Molly nodded along until the timeline reached the bottom corner, where a small photo had been pasted: a figure in a rainbow wig, frozen mid-balloon-twist, standing three feet from the loot. He appeared in every photograph, like a haunting. In one, he looked directly into the lens. "There's a clown." Bill went very still. "You saw him." "He's doing balloon animals next to the presents in every single picture." "That's Chuckles the Magnificent. Formerly Chuck the Quiet, of the Brantford crew. Never cross a man who's juggled on your corner, Molly. He knows things." She leaned back against the van wall and closed her eyes, the way she did before agreeing to things she would later regret. The plan was insane. The plan was beneath them. The plan also involved a slide where Bill had calculated their take in Nerf blasters per minute, and some traitorous part of her brain had found that impressive. "Fine. Nobody makes eye contact with the clown." Bill offered his fist, misty with emotion. "This time tomorrow, we're rich in Nerf." Molly bumped it. "If one child cries, we abort." He nodded solemnly, then clicked to the final slide: a selfie of himself outside a Party City at midnight, beaming, holding a bulging shopping bag. Molly looked at the bag, then at him, then at the bag again. "Please tell me those aren't clown noses." "We can discuss it on the drive."
created_at
"2026-09-02T09:35:34.57Z"
9/2/2026, 9:35:34 AM
model
"z-ai/glm-5.3-flash-20260826"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
4030
moderation_latency
(null)
generation_time
286537
tokens_prompt
75
tokens_completion
11130
native_tokens_prompt
73
native_tokens_completion
12414
native_tokens_completion_images
(null)
native_tokens_reasoning
11730
native_tokens_cached
0
num_media_prompt
(null)
num_input_audio_prompt
(null)
num_media_completion
0
num_search_results
(null)
num_fetches
(null)
web_search_engine
(null)
origin
"https://ncbench.com/"
is_byok
false
finish_reason
"stop"
native_finish_reason
"stop"
service_tier
(null)
usage
0.003108975
router
(null)
user_agent
"langchainjs-openai/1.0.0 ((node/v24.19.0; linux; x64))"
http_referer
(null)
request_id
"req-1788341734-TIwxlDonqh0O1UkbCeM9"
session_id
(null)
preset_id
(null)
api_type
"completions"
id
"gen-1788341734-dg6MQqoZZ6P3D7ucjRqz"
upstream_id
"20260902173534bc35a3f64a0443db"
provider_responses
0
endpoint_id
"8e9fe48b-2f91-41c3-a8a7-e4a93a8c4ff0"
id
"20260902173534bc35a3f64a0443db"
is_byok
false
latency
4030
model_permaslug
"z-ai/glm-5.3-flash-20260826"
provider_name
"Z.AI"
status
200
total_cost
0.003108975
cache_discount
(null)
upstream_inference_cost
0
provider_name
"Z.AI"
response_cache_source_id
(null)
data_region
"global"
workspace_id
"97e315e5-d303-487d-83c1-83180e8a13d4"
Evaluation details
Result
Evaluator
Details
Meta Data
100.00%
Matches word count
n/a
neededClean
false
words
499
95.31%
Dialogue to Total Word Ratio
Ratio: 29.02%, Deviation: 0.98%
neededClean
false
wordsTotal
510
wordsDialogue
148
97.6534%