Mistral: Mistral Small 3

mistralai/mistral-small-24b-instruct-2501 · Mistral AIPartial agreement

Tokenizer lineage (measured): Cognitivecomputations
Evidence confidence: Moderate
Declared tokenizer tag: Mistral
Entered catalogue: 30 January 2025
Last tested: 21 August 2026
Measurement: Measured

Evidence

exact tokenizer signature shared with 1 model

Compare with Venice: Uncensored

Download JSON Report a correction

Identity statement

Mistral: Mistral Small 3 is Mistral AI's model. Its measured tokenizer sits with the Cognitivecomputations group, which is evidence that it builds on the Cognitivecomputations tokenizer. Tokenizer reuse is normal engineering, open vocabularies travel between labs, and this says nothing against Mistral AI's authorship of the model itself.

Evidence stack

Signals of different kinds, weighed together

Signal	Result	Reference
Tokenizer signature	Exact match with 1 other model	tk_ddc193e9
API surface	No other entry shares this exact contract	api_85e3f67c
Context and output	32,768 context · 16,384 max output	declared
Reasoning contract	None declared	declared
Serving providers observed	DeepInfra	measured 21 August 2026

Shares this fingerprint

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

Venice: Uncensoredcognitivecomputations/dolphin-mistral-24b-venice-editionexact · 50/50

Closest measured models

Agreement across the mutually clean test strings

Venice: Uncensoredcognitivecomputations/dolphin-mistral-24b-venice-edition50/50 · 100%TheDrummer: Cydonia 24B V4.1thedrummer/cydonia-24b-v4.149/49 · 100%Mistral: Mistral Nemomistralai/mistral-nemo48/50 · 96%NVIDIA: Nemotron 3 Nano 30B A3Bnvidia/nemotron-3-nano-30b-a3b48/50 · 96%NVIDIA: Nemotron 3.5 Lightningnvidia/nemotron-3.5-lightning48/50 · 96%NVIDIA: Nemotron 3 Supernvidia/nemotron-3-super-120b-a12b48/50 · 96%

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-long18

spaces-203

spaces-603

tabs-204

newlines-207

mixed-ws7

Digits

digits-99

digits-1212

digits-3030

digits-sep12

float-long22

CJK

zh-common16

zh-long20

zh-rare20

ja-kana11

ja-kanji13

ko13

Other scripts

ru12

ar9

he10

hi17

th18

el13

Emoji

emoji-basic20

emoji-skin40

emoji-zwj-family19

emoji-zwj-x357

emoji-flags32

emoji-prof35

Rare Unicode

math33

boxdraw30

combining14

cjk-ext-b16

surrogates53

zalgo36

rtl-mix6

Code

py-code25

py-indent15

json25

html16

regex49

camel6

snake7

Repetition and encodings

rare-word-x536

repeat-tok41

base6439

hex19

url19

uuid36

An x marks a row excluded as corrupt (caching or provider interference detected during measurement). Overhead subtracted: 163 prompt tokens.

Declared record

What the catalogue claims about this model

Mistral Small 3 is a 24B-parameter language model optimized for low-latency performance across common AI tasks. Released under the Apache 2.0 license, it features both pre-trained and instruction-tuned versions designed...

Modalities: text->text
Declared tokenizer: Mistral
Prompt price: $0.05 / M tokens
Completion price: $0.08 / M tokens

Supported parameters

frequency_penaltylogit_biasmax_tokensmin_ppresence_penaltyrepetition_penaltyresponse_formatseedstopstructured_outputstemperaturetop_ktop_p

defaults: {"temperature":0.3,"top_p":null,"frequency_penalty":null}

Catalogue entry Weights on Hugging Face

History

Every observation, kept as taken

30 January 2025Enters the OpenRouter catalogue declared as Mistral.
21 August 2026Fingerprinted in the YFarmX catalogue sweep · 50 of 50 strings measured clean.
21 August 2026YFarmX assessment: consistent with the Cognitivecomputations family, moderate confidence.

Cite this page as the evidence record for Mistral: Mistral Small 3: 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.