AI Desk · Model Identity · Compare

Mistral: Mistral Small 4 · NVIDIA: Nemotron 3 Nano Omni (free)

Identity similarity: highNear-exact agreement · 48/50Agreement this close usually means one tokenizer behind two serving templates, or a small vocabulary revision inside one family. The differing rows below say which strings part.

Mistral: Mistral Small 4 recordNVIDIA: Nemotron 3 Nano Omni (free) recordOpen in the interactive bench

Tokenizer fingerprint

48 of 50 mutually clean strings agree
Test stringMistral: Mistral Small 4NVIDIA: Nemotron 3 Nano Omni (free)Verdict
en-prose1414match
en-long1818match
spaces-2033match
spaces-6033match
tabs-2044match
newlines-2077match
mixed-ws77match
digits-999match
digits-121212match
digits-303030match
digits-sep1212match
float-long2222match
zh-common1616match
zh-long2020match
zh-rare2020match
ja-kana1111match
ja-kanji1313match
ko1313match
ru1212match
ar99match
he1010match
hi1414match
th1717match
el1313match
emoji-basic2020match
emoji-skin4040match
emoji-zwj-family1919match
emoji-zwj-x35757match
emoji-flags3232match
emoji-prof3535match
math3333match
boxdraw3030match
combining1414match
cjk-ext-b1616match
surrogates5353match
zalgo3636match
rtl-mix66match
py-code2525match
py-indent1515match
json2525match
html1616match
regex4949match
camel66match
snake77match
rare-word-x53635differs
repeat-tok4140differs
base643939match
hex1919match
url1919match
uuid3636match

Values are the marginal prompt-token cost of each string with the model's fixed overhead subtracted. An x marks a row excluded as corrupt for that model.

API surface

The declared serving contracts differ
FieldMistral: Mistral Small 4NVIDIA: Nemotron 3 Nano Omni (free)Verdict
Context length262144256000× differs
Max output tokens65536× differs
Supported parametersfrequency_penalty, include_reasoning, max_tokens, presence_penalty, reasoning, reasoning_effort, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_pinclude_reasoning, max_tokens, reasoning, seed, temperature, tool_choice, tools, top_p× differs
Defaults{"temperature":null,"top_p":null,"top_k":null,"frequency_penalty":null,"presence_penalty":null,"repetition_penalty":null}{"temperature":0.6,"top_p":0.95,"top_k":null,"frequency_penalty":null,"presence_penalty":null,"repetition_penalty":null}× differs
Reasoning contract{"mandatory":false,"default_enabled":false,"supported_efforts":["high","none"],"default_effort":"high"}{"mandatory":false,"default_enabled":true,"supports_max_tokens":true}× differs
Modalitiestext+image->texttext+image+audio+video->text× differs

Method: tokenizer fingerprinting, v1, 50 strings. Raw data on the data page. This page is static and dated; the interactive bench compares any two of the 422 catalogue models.