DeepSeek: DeepSeek V3.2 Exp

deepseek/deepseek-v3.2-exp · DeepSeekDeclared and measured agree

Tokenizer group (measured): DeepSeek
Evidence confidence: High
Declared tokenizer tag: DeepSeek
Entered catalogue: 29 September 2025
Last tested: 21 August 2026
Measurement: Measured

Evidence

exact tokenizer signature shared with 9 models

Compare with DeepSeek: DeepSeek V4 Pro 0813

Download JSON Report a correction

Identity statement

DeepSeek: DeepSeek V3.2 Exp is DeepSeek's model, and its measured tokenizer sits in DeepSeek's own signature group alongside its sibling models. Measured models from StepFun carry the same vocabulary, which reads as reuse of DeepSeek's tokenizer line and questions nobody's authorship; they are listed under "Shares this fingerprint" below.

Evidence stack

Signals of different kinds, weighed together

Signal	Result	Reference
Tokenizer signature	Exact match with 9 other models	tk_73c4f0dc
API surface	No other entry shares this exact contract	api_e35de931
Context and output	163,840 context · 65,536 max output	declared
Reasoning contract	Optional	declared
Serving providers observed	Novita	measured 21 August 2026

Shares this fingerprint

Exact signature first; template-boundary shifts of the same signature beneath

DeepSeek: DeepSeek V3deepseek/deepseek-chatexact · 50/50 DeepSeek: DeepSeek V3 0324deepseek/deepseek-chat-v3-0324exact · 50/50 DeepSeek: DeepSeek V3.1 Terminusdeepseek/deepseek-v3.1-terminusexact · 50/50 DeepSeek: DeepSeek V3.2deepseek/deepseek-v3.2exact · 50/50 DeepSeek: DeepSeek V4 Flash 0423deepseek/deepseek-v4-flashexact · 50/50 DeepSeek: DeepSeek V4 Flash 0731deepseek/deepseek-v4-flash-0731exact · 50/50 DeepSeek: DeepSeek V4 Pro 0813deepseek/deepseek-v4-pro-0813exact · 50/50 StepFun: Step 3.5 Flashstepfun/step-3.5-flashexact · 50/50 StepFun: Step 3.7 Flashstepfun/step-3.7-flashexact · 50/50 DeepSeek: DeepSeek V3.1deepseek/deepseek-chat-v3.1same tokenizer · template shift DeepSeek: R1 Distill Llama 70Bdeepseek/deepseek-r1-distill-llama-70bsame tokenizer · template shift

Closest measured models

Agreement across the mutually clean test strings

DeepSeek: DeepSeek V4 Pro 0813deepseek/deepseek-v4-pro-081350/50 · 100%DeepSeek: DeepSeek V4 Flash 0731deepseek/deepseek-v4-flash-073150/50 · 100%DeepSeek: DeepSeek V4 Flash 0423deepseek/deepseek-v4-flash50/50 · 100%StepFun: Step 3.5 Flashstepfun/step-3.5-flash50/50 · 100%StepFun: Step 3.7 Flashstepfun/step-3.7-flash50/50 · 100%DeepSeek: DeepSeek V3.2deepseek/deepseek-v3.250/50 · 100%

Click a row to open the full comparison.

The measured fingerprint

Marginal prompt-token cost of each test string, grouped by script

All 50 rows

English and whitespace

en-prose14

en-long16

spaces-203

spaces-603

tabs-204

newlines-204

mixed-ws7

Digits

digits-93

digits-124

digits-3010

digits-sep7

float-long9

CJK

zh-common9

zh-long10

zh-rare18

ja-kana9

ja-kanji11

ko13

Other scripts

ru13

ar10

he14

hi18

th11

el16

Emoji

emoji-basic10

emoji-skin14

emoji-zwj-family11

emoji-zwj-x333

emoji-flags16

emoji-prof19

Rare Unicode

math22

boxdraw27

combining10

cjk-ext-b16

surrogates33

zalgo36

rtl-mix6

Code

py-code25

py-indent15

json20

html16

regex47

camel7

snake8

Repetition and encodings

rare-word-x526

repeat-tok41

base6434

hex12

url17

uuid27

An x marks a row excluded as corrupt (caching or provider interference detected during measurement). Overhead subtracted: 4 prompt tokens.

Declared record

What the catalogue claims about this model

DeepSeek-V3.2-Exp is an experimental large language model released by DeepSeek as an intermediate step between V3.1 and future architectures. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism...

Modalities: text->text
Declared tokenizer: DeepSeek
Prompt price: $0.27 / M tokens
Completion price: $0.41 / M tokens

Supported parameters

frequency_penaltyinclude_reasoninglogit_biaslogprobsmax_tokensmin_ppresence_penaltyreasoningrepetition_penaltyresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_ktop_logprobstop_p

defaults: {"temperature":0.6,"top_p":0.95,"frequency_penalty":null}

Catalogue entry Weights on Hugging Face

History

Every observation, kept as taken

29 September 2025Enters the OpenRouter catalogue declared as DeepSeek.
21 August 2026Fingerprinted in the YFarmX catalogue sweep · 50 of 50 strings measured clean.
21 August 2026YFarmX assessment: consistent with the DeepSeek family, high confidence.

Cite this page as the evidence record for DeepSeek: DeepSeek V3.2 Exp: the URL is stable, measurements are dated, and revisions append to the history above. Method: tokenizer fingerprinting, v1, 50 strings. The raw data behind every figure is on the data page.