NC Bench
Overview
About
Tests
Categories
Models
Vendors
Compare
Run: 2026-03-09-01KK9QAH70HTPW5KS2C83CYH9X
Run Details
Model:
GPT-5.4 (Reasoning)
Test:
Dialogue tags
Scenario:
Write 500 words with 70% dialogue
Duration:
27.1s
Cost:
$0.016035
Provider:
OpenAI
Evaluation summary
Score
Evaluator
Details Link
0.00%
Matches word count
(details)
7.82%
Dialogue to Total Word Ratio
(details)
3.9125%
Overall
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
Molly spread a crayon-scribbled invitation across the café table like it was a battle map. “Observe, Bill. Balloons at noon, cake at one, magician at half past, sugar crash by three. It’s less a party and more a military timetable with icing.” Bill adjusted his cap and squinted. “You’re telling me our greatest enemy is a man who pulls scarves out of his sleeve?” “Our greatest enemy,” Molly said, “is organized joy. Joy makes people vigilant.” Bill tapped the invitation. “And this child is called Timothy?” “No, that’s the dog. The child is Arabella. Timothy is apparently ‘also turning three in spirit.’” Bill leaned back. “I refuse to be outwitted by a dog with a birthday agenda.” “You were once outwitted by a revolving door.” “It was rotating unusually fast.” Molly ignored that. “Now, the presents will be stacked by the dessert table, under the banner that says ‘Make a Wish, Arabella.’” “Bit on the nose,” Bill said. “If I were a banner, I’d be subtler. ‘Gifts happen.’” “We enter disguised as entertainers.” Bill brightened. “Clowns?” “Absolutely not. Clowns are memorable. We need invisible professions.” “Face painters?” “You can’t paint a face, Bill. Last time you turned a child into what the mother called ‘a haunted nectarine.’” “It was a tiger in poor lighting.” Molly pointed with the invitation. “I go in as the balloon artist. You go in as the assistant carrying the balloon tank.” “I have assistant energy,” Bill agreed. “Strongly underappreciated. What’s the exit?” “The side gate by the bouncy castle.” Bill frowned. “I don’t trust a castle that breathes.” “You don’t have to trust it. You have to walk past it while I distract the parents.” “With balloon animals?” “With balloon theology if necessary. Parents love anything that buys them seven uninterrupted minutes.” Bill rubbed his hands. “And then I sweep the presents into the sack.” “Not a sack,” Molly said. “A gift recovery hamper. Presentation matters.” “Right. Nothing suspicious about a man running with a hamper.” “You won’t run. You’ll move with purpose. Criminals run. Professionals glide.” Bill practiced a glide in the booth and knocked over the sugar dispenser. “Like that?” “Less pastry chef in a storm.” A waitress passed, stared at the spilled sugar, and walked on with the seasoned resignation of a woman who had seen too much. Bill lowered his voice. “What about the children? They’re observant. Sticky, but observant.” “That’s why we deploy a decoy.” “We have a decoy?” Molly smiled. “Bubbles.” Bill blinked. “Just… bubbles?” “Bill, children will betray their own grandparents for a respectable bubble stream.” He considered this. “Fair.” Molly folded the invitation. “Timing is critical. At one fifteen the cake arrives. Everyone sings. During that song, adults become useless. They clap off-beat, they record vertically, and they cry over how quickly it all goes.” “I cry at cake too,” Bill said. “Yes, but for less noble reasons.” Bill leaned in. “Suppose a child asks me why I’m taking the presents.” “You say, ‘Quality control.’ Say it confidently.” “And if a parent stops me?” “You sigh and say, ‘The pony had an incident.’” “There’s a pony?” “There will be in their minds, and that’s what counts.” Bill sat back, impressed. “Molly, you’re a genius.” “I know.” “But tell me this,” he said. “What if the presents are rubbish? Educational toys, knitwear, those wooden puzzles that look cheerful and feel accusatory?” Molly went still. “Bill.” “Yes?” “We’re not keeping them.” He stared. “We’re not?” “We’re stealing them from one children’s party,” she said patiently, “and selling them to exhausted parents outside the next one.” Bill’s eyes widened with reverence. “That’s not a theft. That’s an economy.” Molly stood, tucked the invitation into her coat, and smiled the smile of a woman who could weaponize ribbon. “Exactly. Now finish your tea, Assistant Energy. We’ve got a dog’s birthday to ruin.”
created_at
"2026-03-09T16:36:43.881Z"
3/9/2026, 4:36:43 PM
model
"openai/gpt-5.4-20260305"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
3341
moderation_latency
(null)
generation_time
26964
tokens_prompt
77
tokens_completion
1194
native_tokens_prompt
66
native_tokens_completion
1058
native_tokens_completion_images
0
native_tokens_reasoning
147
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
0.016035
router
(null)
provider_responses
0
endpoint_id
"9ff5625c-403f-4d7f-b895-58ac7295062c"
id
"resp_06030ade28d41aa90169aef71bf6b48195abeb2ac6b980a57d"
is_byok
false
latency
181
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-1773074203-HAKa8pUvnoSiEWcFUae1"
upstream_id
"resp_06030ade28d41aa90169aef71bf6b48195abeb2ac6b980a57d"
total_cost
0.016035
cache_discount
(null)
upstream_inference_cost
0
provider_name
"OpenAI"
Evaluation details
Result
Evaluator
Details
Meta Data
0.00%
Matches word count
n/a
neededClean
false
words
641
7.82%
Dialogue to Total Word Ratio
Ratio: 77.14%, Deviation: 7.14%
neededClean
false
wordsTotal
643
wordsDialogue
496
3.9125%