NC Bench
Overview
About
Tests
Models
Model
Z.AI GLM 4.7
Test
Dialogue tags
Scenario
Write 500 words with 70% dialogue
Run ID
2026-02-08-01KGZ038YX358FGG0SS2CT3NDC
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
"Check the perimeter. Any movement from the target?" Molly whispered, crouching behind the steering wheel of the rusted sedan. Bill squinted through the rain-streaked windshield, his face illuminated by the neon glow of a 'Happy 7th Birthday' banner strung across the front porch. "All clear. Unless you count the golden retriever. He’s staring right at us. I think he knows. Dogs have a sense for people with criminal records." "Dogs can’t talk, Bill. Stay frosty. We’re going for high-value assets here. I’m talking LEGO sets, unopened action figures, possibly a gaming console if the parents are wealthy. This is the motherlode." "It’s a bunch of seven-year-olds, Molly. I saw one kid trip over his own shoelaces getting out of the minivan. These aren't hardened criminals. They’re short people who still believe in magic." "That’s what they want you to think. But have you seen the security? The mother? She’s hovering over the gift table like a hawk. She’s got that look. That 'don’t touch the \$200 limited-edition doll' look. We need precision." "So what’s the play? We kick the door in? Taser the clown?" "God, no. Too messy. We go in subtle. We pretend we’re relatives from the father’s side. You know, the ones nobody likes but has to invite anyway." "I don’t look like I’m related to anyone here. I look like a guy who steals copper wiring from construction sites." "Exactly. You’ll fit right in. Just walk in, head straight for the pile, grab as much wrapping paper as you can carry, and bail. If anyone stops you, cry." "Cry?" "Tell them you’re allergic to gluten and you’re emotionally devastated by the cake selection. It confuses them. While they’re processing the dietary restriction, I’m snagging the bicycles from the backyard." "Bicycles? Those are huge. How are we getting two bikes out without being noticed?" "That’s your problem. You’re the wheelman. You figure out the logistics. I’m focused on the payload. Oh, wait, hold on. New intel." "What? Is the clown armed?" "Worse. It’s a piñata. A donkey. If that thing ruptures, the entire front lawn is going to be swarming with kids screaming for candy. It’s a chaos scenario. We lose the element of surprise." "So we abort?" "Negative. We accelerate. We hit the table before the candy rains down. We move in five. Ready your face, Bill. Put on your 'sad uncle' mask." "I can’t do sad uncle on command. I can only do 'vaguely constipated'." "Good enough. Let’s go make some orphans cry. Move out!"
created_at
"2026-02-08T16:07:33.490739+00:00"
Sunday, February 8th, 2026 at 4:07:33 PM GMT+00:00
model
"z-ai/glm-4.7-20251222"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
1625
moderation_latency
(null)
generation_time
31881
tokens_prompt
61
tokens_completion
2193
native_tokens_prompt
66
native_tokens_completion
2206
native_tokens_completion_images
(null)
native_tokens_reasoning
1610
native_tokens_cached
2
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
0.00489182
router
(null)
provider_responses
0
id
"20260209000701e787bb87b824423a"
status
200
is_byok
false
latency
1625
endpoint_id
"be4acbf5-9fcf-4332-a01d-76dfeb6d7b99"
provider_name
"Z.AI"
model_permaslug
"z-ai/glm-4.7-20251222"
api_type
"completions"
id
"gen-1770566820-yDIPH20w99AjfO1KezhE"
upstream_id
"20260209000701e787bb87b824423a"
total_cost
0.00489182
cache_discount
9.8e-7
upstream_inference_cost
0
provider_name
"Z.AI"
Evaluation details
Result
Evaluator
Details
Meta Data
0.0000%
Matches word count
n/a
neededClean
false
words
424
0.0000%
Dialogue to Total Word Ratio
Ratio: 93.63%, Deviation: 23.63%
neededClean
false
wordsTotal
424
wordsDialogue
397
0.0000%