AI Desk · Model Identity · Compare
NVIDIA: Nemotron 3.5 Content Safety (free) · Xiaomi: MiMo-V2.5-Pro
Identity similarity: lowDivergent fingerprints · 6/35These models tokenise differently. Divergence on the high-information strings (CJK, emoji, digits) separates unrelated vocabularies within a handful of samples.
NVIDIA: Nemotron 3.5 Content Safety (free) recordXiaomi: MiMo-V2.5-Pro recordOpen in the interactive bench
Tokenizer fingerprint
6 of 35 mutually clean strings agree| Test string | NVIDIA: Nemotron 3.5 Content Safety (free) | Xiaomi: MiMo-V2.5-Pro | Verdict |
|---|---|---|---|
| en-prose | 14 | 12 | differs |
| en-long | 15 | 14 | differs |
| spaces-20 | 2 | 1 | differs |
| spaces-60 | 2 | 1 | differs |
| tabs-20 | 2 | x | – |
| newlines-20 | 2 | x | – |
| mixed-ws | 2 | 3 | differs |
| digits-9 | 10 | 7 | differs |
| digits-12 | 13 | x | – |
| digits-30 | 31 | 28 | differs |
| digits-sep | 13 | 10 | differs |
| float-long | 23 | 20 | differs |
| zh-common | 11 | 8 | differs |
| zh-long | 14 | x | – |
| zh-rare | 26 | 15 | differs |
| ja-kana | 7 | 6 | differs |
| ja-kanji | 10 | 11 | differs |
| ko | 11 | 15 | differs |
| ru | 12 | 14 | differs |
| ar | 9 | 6 | differs |
| he | 12 | 7 | differs |
| hi | 8 | 27 | differs |
| th | 10 | 10 | match |
| el | 13 | x | – |
| emoji-basic | 5 | x | – |
| emoji-skin | 10 | 8 | differs |
| emoji-zwj-family | 8 | 8 | match |
| emoji-zwj-x3 | 22 | x | – |
| emoji-flags | 9 | 6 | differs |
| emoji-prof | 12 | x | – |
| math | 26 | 24 | differs |
| boxdraw | 18 | x | – |
| combining | 10 | 3 | differs |
| cjk-ext-b | 17 | 10 | differs |
| surrogates | 28 | x | – |
| zalgo | 22 | 34 | differs |
| rtl-mix | 5 | 5 | match |
| py-code | 26 | 23 | differs |
| py-indent | 12 | 13 | differs |
| json | 28 | x | – |
| html | 14 | 14 | match |
| regex | 44 | x | – |
| camel | 6 | 3 | differs |
| snake | 11 | x | – |
| rare-word-x5 | 25 | 29 | differs |
| repeat-tok | 20 | 20 | match |
| base64 | 34 | 34 | match |
| hex | 16 | x | – |
| url | 23 | x | – |
| uuid | 37 | 34 | differs |
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| Field | NVIDIA: Nemotron 3.5 Content Safety (free) | Xiaomi: MiMo-V2.5-Pro | Verdict |
|---|---|---|---|
| Context length | 128000 | 1050000 | × differs |
| Max output tokens | 8192 | 131072 | × differs |
| Supported parameters | include_reasoning, max_tokens, reasoning, seed, temperature, top_p | frequency_penalty, include_reasoning, logit_bias, logprobs, max_tokens, min_p, presence_penalty, reasoning, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, top_p | × differs |
| Defaults | {"temperature":null,"top_p":null,"top_k":null,"frequency_penalty":null,"presence_penalty":null,"repetition_penalty":null} | {"temperature":1,"top_p":0.95,"top_k":null,"frequency_penalty":null,"presence_penalty":null,"repetition_penalty":null} | × differs |
| Reasoning contract | {"mandatory":false,"default_enabled":true} | {"mandatory":false} | × differs |
| Modalities | text+image->text | text->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.