NC Bench
Overview
About
Tests
Categories
Models
Vendors
Compare
Run: 2026-04-25-01KQ2HR7K3JKDQ1DPY5WV08MVB
Run Details
Model:
DeepSeek V4 Flash
Test:
Dialogue tags
Scenario:
Write 500 words with 30% dialogue
Duration:
22.2s
Cost:
$0.000222
Provider:
DeepSeek
Evaluation summary
Score
Evaluator
Details Link
86.38%
Matches word count
(details)
0.00%
Dialogue to Total Word Ratio
(details)
43.1902%
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 fairy lights strung across the community hall ceiling twitched like nervous fireflies, casting a jaunty, multi-coloured glow on Molly’s furrowed brow. She leaned over the blueprints—a crayon drawing on a napkin that Bill called a “reconnaissance sketch.” It was mostly jammy handprints and a single, shaky arrow labelled “Prize Table.” “You’re telling me,” Molly whispered, her voice a low rumble that shook the glitter on the tablecloth, “that our target is a six-year-old’s pirate-themed birthday party.” Bill adjusted his balaclava, which was, against all professional protocol, neon orange. “Correct. The payload is significant, Molly. We’re talking a light-saber that actually makes the swoosh noise, a unicorn that poops sequins, and a three-foot-tall Lego pirate ship that’s eighty percent assembled. The parents are still buying wine in the lobby. We have a three-minute window during the Pin the Parrot on the Pirate game.” “We are professional art thieves,” Molly hissed, jabbing a finger at the napkin. “We’ve done a midnight swap of a Monet in Prague. We’ve got a guy on retainer who can forge a Rembrandt’s signature. And you want us to steal a light-saber that goes swoosh?” “The Monet was for insurance fraud,” Bill said, taking a deep, philosophical breath. “This is for joy. Real, unadulterated, sweaty-palmed, sugar-rush joy. And for resale. The unicorn sequin pooper is going for four hundred on eBay.” Molly pinched the bridge of her nose. “And the plan? The one that involves my borrowed trench coat, a balloon animal, and a carpet of screaming children?” “Stage one,” Bill said, ticking off on his gloved fingers. “You, dressed as a pirate, disrupt the game by ‘accidentally’ popping the parrot with your peg-leg. Chaos ensues. I, dressed as a confused clown, will use a high-pitched squeaky hammer to herd the weeping ankle-biters toward the cake table. You then sweep the table, gifting the haul to me via the ‘sick child’ decoy in the duffel bag by the fire exit. It’s flawless.” Molly stared at him. “You want me to wear a peg-leg.” “And an eye patch. It’s thematic.” “We don’t have a sick child.” “We have a tawny cushion that looks vaguely human if you squint and hold it at speed.” A small, sticky hand tugged on Molly’s trench coat. A boy with a frosting mustache looked up at her with unnerving clarity. “Are you here for the presents?” he asked. Molly froze. Bill’s balaclava slipped, revealing a grin that was far too eager. “No, champ,” she said, her voice a croak. “We’re the entertainment.” The boy squinted. “Your peg-leg is on the wrong leg.” Molly looked down. The foam rubber stump was, indeed, strapped to her left leg instead of her right. A high-pitched squeak came from Bill’s direction. It was not the hammer. It was his nervous bladder. Thirty seconds later, the disco music for “Pin the Parrot” kicked in, and a hurricane of glitter and shrieking descended. The plan was already dead. The light-saber, however, was right there, on the table. Molly sighed. Some jobs just take a different kind of professional.
created_at
"2026-04-25T14:47:27.591Z"
4/25/2026, 2:47:27 PM
model
"deepseek/deepseek-v4-flash-20260423"
app_id
182717
external_user
(null)
streamed
true
cancelled
false
latency
638
moderation_latency
(null)
generation_time
11489
tokens_prompt
77
tokens_completion
836
native_tokens_prompt
67
native_tokens_completion
758
native_tokens_completion_images
(null)
native_tokens_reasoning
0
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
0.00022162
router
(null)
provider_responses
0
endpoint_id
"934a69f9-bd54-474b-beca-24560f721e12"
is_byok
false
latency
52
model_permaslug
"deepseek/deepseek-v4-flash-20260423"
provider_name
"DeepInfra"
status
(null)
1
endpoint_id
"722e0746-2034-4d6e-a08f-e6271849ee9e"
id
"0a8d5178-e578-4d2a-ae43-bf267b931927"
is_byok
false
latency
274
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-1777128447-Ki4Z6xKZjDFYBlhX89i6"
session_id
(null)
api_type
"completions"
id
"gen-1777128447-4IMChQPhSWtBEfqp5L3k"
upstream_id
"0a8d5178-e578-4d2a-ae43-bf267b931927"
total_cost
0.00022162
cache_discount
(null)
upstream_inference_cost
0
provider_name
"DeepSeek"
response_cache_source_id
(null)
Evaluation details
Result
Evaluator
Details
Meta Data
86.38%
Matches word count
n/a
neededClean
false
words
511
0.00%
Dialogue to Total Word Ratio
Ratio: 55.87%, Deviation: 25.87%
neededClean
false
wordsTotal
528
wordsDialogue
295
43.1902%