NC Bench
Overview
About
Tests
Categories
Models
Vendors
Compare
Run: 2026-03-09-01KKAC5EM6766KDXJGR651MSYJ
Run Details
Model:
GPT-5.4 (Reasoning, Low)
Test:
Dialogue tags
Scenario:
Write 200 words with 10% dialogue
Duration:
13.1s
Cost:
$0.008145
Provider:
OpenAI
Evaluation summary
Score
Evaluator
Details Link
35.00%
Matches word count
(details)
4.74%
Dialogue to Total Word Ratio
(details)
19.8711%
Overall
Model input
User/Human
You are a author of comedic heist books. Please write a 200 word scene that has 10% 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
Molly spread the blueprint across a sticky café table, smoothing it between a sugar jar and a plate of exhausted biscuits. The target was not a bank, museum, or diamond exchange, but a semi-detached house on Larkspur Road where six-year-old Oliver Finch would soon celebrate with twenty shrieking guests, two distracted parents, and a mountain of ribboned loot stacked beside a rented magician. Outside, rain slicked the pavement; inside, Bill wore the expression of a man being asked to burgle innocence itself, provided innocence came with batteries included. “Clowns,” Molly whispered, “never frisk anyone.” “Good,” Bill said. “I bruise like a peach.” She tapped the sketch: garden gate, bouncy castle, kitchen, present table. “Entry during pass-the-parcel. Exit during cake panic. The mother will be slicing, the father will be apologizing, and the magician will be extracting a hamster from somewhere legally ambiguous.” Bill adjusted his bow tie, borrowed from a waiter and still smelling faintly of soup. The plan was absurd, which to Molly meant nearly perfect. He would juggle by the patio, drawing the children like moths to a tax-deductible flame, while she swept the presents into a laundry sack marked CHARITY. “Then smile,” Molly said, “and keep the confetti out of your lock picks.” Bill nodded gravely, as if professionalism still mattered under a balloon arch.
created_at
"2026-03-09T22:40:57.485Z"
3/9/2026, 10:40:57 PM
model
"openai/gpt-5.4-20260305"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
5350
moderation_latency
(null)
generation_time
13043
tokens_prompt
77
tokens_completion
368
native_tokens_prompt
66
native_tokens_completion
532
native_tokens_completion_images
0
native_tokens_reasoning
232
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
"completed"
usage
0.008145
router
(null)
provider_responses
0
endpoint_id
"9ff5625c-403f-4d7f-b895-58ac7295062c"
id
"resp_06d89afcf13af20e0169af4c798a948190a3c1a027f7dce600"
is_byok
false
latency
207
model_permaslug
"openai/gpt-5.4-20260305"
provider_name
"OpenAI"
status
200
user_agent
"langchainjs-openai/1.0.0 ((node/v24.14.0; linux; x64))"
http_referer
(null)
api_type
"completions"
id
"gen-1773096057-sryf6LZrnXM3SVzusmXP"
upstream_id
"resp_06d89afcf13af20e0169af4c798a948190a3c1a027f7dce600"
total_cost
0.008145
cache_discount
(null)
upstream_inference_cost
0
provider_name
"OpenAI"
Evaluation details
Result
Evaluator
Details
Meta Data
35.00%
Matches word count
n/a
neededClean
false
words
218
4.74%
Dialogue to Total Word Ratio
Ratio: 23.21%, Deviation: 13.21%
neededClean
false
wordsTotal
224
wordsDialogue
52
19.8711%