NC Bench
Overview
About
Tests
Categories
Models
Vendors
Compare
Run: 2026-04-25-01KQ2H4V24QV8X2D3W95WMAD1X
Run Details
Model:
DeepSeek V4 Flash (Reasoning)
Test:
Dialogue tags
Scenario:
Write 500 words with 70% dialogue
Duration:
14.2s
Cost:
$0.000268
Provider:
DeepSeek
Evaluation summary
Score
Evaluator
Details Link
0.00%
Matches word count
(details)
0.02%
Dialogue to Total Word Ratio
(details)
0.0083%
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
“You can’t be serious.” Bill’s whisper was a rusty hinge, scraping against the back of a minivan they’d tucked themselves behind. “A children’s party? We’re master thieves, Molly. We’ve lifted a Fabergé egg from a Russian oligarch’s yacht. We’ve nabbed the Mona Lisa’s cousin—the one in the basement of the Louvre.” “That was a print,” Molly said, squinting through a pair of binoculars at the inflatable bouncy castle across the lawn. “And the egg was a chocolate replica. The real one was in the gift shop. But this—this is solid gold, Bill. Look at those gift bags. They’re not filled with plastic whistles. That’s a Tiffany ribbon. I can smell the profit.” “You can’t smell profit.” “I can smell merino wool and envy. The party favors are from Saks. The birthday boy’s mother is a hedge-fund wife who thinks a piñata is a type of tapas. Those presents under the table? There’s a PS5, an iPad, and—I swear on my lockpicks—a signed first edition of Harry Potter and the Philosopher’s Stone.” Bill rubbed his face. “We’d have to get past twenty-seven five-year-olds. They’re vicious. They bite. And they scream. Remember the security guard at the museum? That was a low, adult scream. A child’s scream is a weapon of mass distraction.” “Which is why we go in as entertainers,” Molly said, pulling a crumpled flyer from her pocket. “Clowning for Cash. I’ve got a unicycle and a balloon swords.” “You cannot ride a unicycle.” “Not yet. But I can wobble menacingly. You’ll be my assistant. We make balloon animals—badly. The kids get distracted. I cry at their disappointment. They laugh. We loot.” Bill stared at her. “You want me to wear face paint?” “And a wig. A red, frizzy one. It’s in the trunk next to the stilts.” “I’m not wearing stilts.” “Fine. You’re the grumpy clown who juggles. I’ve got three tennis balls and a potato. It’s the same shape.” Bill held up a hand. “Let me get this straight. We—two career criminals with a combined thirty-seven arrests—are going to infiltrate a seven-year-old’s birthday party, perform a comedy routine that will traumatize the children, and then make off with their gifts. And you’ve planned this how?” “I’ve got the escape route mapped. Through the hedge, over the fence, past the neighbor’s aggressive poodle. We meet at the ice cream truck.” “The ice cream truck driver is our getaway?” “He’s an ex-con. He owes me a favor. And he has a loud jingle that covers noise.” Bill sighed. “What if the kids like us? What if they want autographs?” “Then we sign ‘Mickey Mouse’ and leave. It’s not rocket science. It’s kindergarten crime.” “And the mother? She’s a hedge-fund wife. She’ll have a security detail. Probably a drone.” “The drone is a Roomba with a camera. I’ve already hacked it. It’s currently set to ‘avoid the bouncy castle and follow the cat.’ We have twelve minutes before it returns to its dock. Twelve minutes to grab, run, and never speak of this again.” Bill looked at the party. He looked at the presents. He looked at his own reflection in the binoculars—a tired face, a thinning hairline, a man who’d once tried to steal a Fabergé egg that was actually chocolate. “Fine,” he said. “But I’m not making balloon dachshunds. I’m drawing the line at dachshunds.” “Deal. Now get the wig. We’ve got a party to ruin.”
created_at
"2026-04-25T14:36:52.167Z"
4/25/2026, 2:36:52 PM
model
"deepseek/deepseek-v4-flash-20260423"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
641
moderation_latency
(null)
generation_time
14124
tokens_prompt
77
tokens_completion
999
native_tokens_prompt
67
native_tokens_completion
923
native_tokens_completion_images
(null)
native_tokens_reasoning
57
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"
usage
0.00026782
router
(null)
provider_responses
0
endpoint_id
"722e0746-2034-4d6e-a08f-e6271849ee9e"
id
"d02808b2-0618-425c-a4d8-bf0f9419d8e6"
is_byok
false
latency
390
model_permaslug
"deepseek/deepseek-v4-flash-20260423"
provider_name
"DeepSeek"
status
200
user_agent
"langchainjs-openai/1.0.0 ((node/v24.14.1; linux; x64))"
http_referer
(null)
request_id
"req-1777127812-jCz25Iqqt8ik8WBl3dha"
session_id
(null)
api_type
"completions"
id
"gen-1777127812-tLYrDhttL8i7p4DgKXqh"
upstream_id
"d02808b2-0618-425c-a4d8-bf0f9419d8e6"
total_cost
0.00026782
cache_discount
(null)
upstream_inference_cost
0
provider_name
"DeepSeek"
response_cache_source_id
(null)
Evaluation details
Result
Evaluator
Details
Meta Data
0.00%
Matches word count
n/a
neededClean
false
words
575
0.02%
Dialogue to Total Word Ratio
Ratio: 83.19%, Deviation: 13.19%
neededClean
false
wordsTotal
583
wordsDialogue
485
0.0083%