NC Bench
Overview
About
Tests
Categories
Models
Vendors
Compare
Run: 2026-09-02-01M1GQSMAXN7ZKJNDXRSPA8EVF
Run Details
Model:
Z.AI GLM 5.3 (Reasoning, Max)
Test:
Dialogue tags
Scenario:
Write 500 words with 30% dialogue
Duration:
11m 8s
Cost:
$0.110516
Provider:
Z.AI
Evaluation summary
Score
Evaluator
Details Link
99.38%
Matches word count
(details)
98.02%
Dialogue to Total Word Ratio
(details)
98.6985%
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 blueprint spread across the kitchen table had been annotated with military precision. Red circles marked the bouncy castle. A dotted line traced the optimal route through the garden, and in one corner, a stern warning about the grandmother had been underlined twice. Molly leaned over the map wearing reading glasses she didn't need, because heists demanded gravitas. Three evenings of surveillance from a parked car, two packets of custard creams, and one deeply suspicious conversation with a postman had gone into this. Little Katy Henderson's seventh birthday would be the score of the decade. "Cake's at three, sugar crash at four, presents at half past," she said, tapping the folding table drawn in the garden's far corner. "Sixteen gifts, including what I believe to be a robot dog that genuinely poops." Bill, her partner of eleven years, was chewing a pen cap down to its molecular components. He had learned long ago to treat her briefings with the reverence of a man casing a casino vault rather than a paddling pool. "Security?" he asked. "A golden retriever named Biscuit and a nan who naps during films." "Armed?" "The nan has a slipper." Bill nodded gravely and made a note. He underlined it twice, which was his personal system for serious threats. Somewhere in his jacket was a laminated card reading REMAIN CALM that he consulted during weddings and trips to the tip. "The real danger is the children," Molly continued. "They're like tiny detectives. One of them once clocked that a magician's dove was just a pigeon with glitter on it. He made a sign." "Saying what?" "Pigeon Fraud." Molly let the words hang in the air the way veterans discuss old wars. "So we go in as entertainers. I'm face painting. You're on balloons." Bill went pale. He had attempted a balloon animal once, alone in his kitchen with the curtains drawn, and the balloon had won. There had been a noise like a gunshot and a red streak across the ceiling that his landlord still asked about. "I can make a snake," he said finally. "Wonderful. You'll make thirty." "Kids notice patterns, Molly. They'll rumble me, and then it's Pigeon Fraud all over again, except this time it's Snake Fraud, and I couldn't survive that." Molly considered this, then reached into her bag and produced a photograph of an ice cream van, sliding it onto the table with the calm confidence of a woman laying down a royal flush. "Ninety-nine flakes for everyone," she said. "While they're hunting sprinkles, we empty the present table and slip out the side gate." "In the ice cream van?" Molly paused. Details were technically Bill's department, and Bill's department was, historically speaking, a catastrophe with a lanyard. Somewhere out there was her brother, who owned a van and still owed her for the incident at the casino, but that was another blueprint, on another kitchen table, for another day. Bill picked up a balloon and began, cautiously, to inflate it. "Snake," he muttered. "Snake. Snake. Snake."
created_at
"2026-09-02T09:39:25.161Z"
9/2/2026, 9:39:25 AM
model
"z-ai/glm-5.3-20260816"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
5923
moderation_latency
(null)
generation_time
667938
tokens_prompt
75
tokens_completion
19728
native_tokens_prompt
73
native_tokens_completion
25094
native_tokens_completion_images
(null)
native_tokens_reasoning
24436
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
0.1105158
router
(null)
user_agent
"langchainjs-openai/1.0.0 ((node/v24.19.0; linux; x64))"
http_referer
(null)
request_id
"req-1788341965-Yvwxk9aslwrfnsHTp1Q4"
session_id
(null)
preset_id
(null)
api_type
"completions"
id
"gen-1788341965-9Hb6kMuc0Q41FcjUJWyx"
upstream_id
"20260902173925594ba241f95d488f"
provider_responses
0
endpoint_id
"db975786-e7f6-4838-9e56-3e0b77f1bdc2"
id
"20260902173925594ba241f95d488f"
is_byok
false
latency
5923
model_permaslug
"z-ai/glm-5.3-20260816"
provider_name
"Z.AI"
status
200
total_cost
0.1105158
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
99.38%
Matches word count
n/a
neededClean
false
words
505
98.02%
Dialogue to Total Word Ratio
Ratio: 30.63%, Deviation: 0.63%
neededClean
false
wordsTotal
506
wordsDialogue
155
98.6985%