AI Desk · Model Identity · Compare

DeepSeek: DeepSeek V4 Pro 0813 · LiquidAI: LFM2.5-2.6B (free)

Identity similarity: lowDivergent fingerprints · 29/50These models tokenise differently. Divergence on the high-information strings (CJK, emoji, digits) separates unrelated vocabularies within a handful of samples.

DeepSeek: DeepSeek V4 Pro 0813 recordLiquidAI: LFM2.5-2.6B (free) recordOpen in the interactive bench

Tokenizer fingerprint

29 of 50 mutually clean strings agree
Test stringDeepSeek: DeepSeek V4 Pro 0813LiquidAI: LFM2.5-2.6B (free)Verdict
en-prose1414match
en-long1618differs
spaces-2033match
spaces-6033match
tabs-2044match
newlines-2044match
mixed-ws77match
digits-933match
digits-1244match
digits-301010match
digits-sep77match
float-long99match
zh-common912differs
zh-long1016differs
zh-rare1818match
ja-kana99match
ja-kanji1111match
ko1311differs
ru1316differs
ar1010match
he1414match
hi1814differs
th1114differs
el1617differs
emoji-basic1010match
emoji-skin1420differs
emoji-zwj-family1111match
emoji-zwj-x33333match
emoji-flags1616match
emoji-prof1919match
math2227differs
boxdraw2729differs
combining1012differs
cjk-ext-b1616match
surrogates3340differs
zalgo3636match
rtl-mix67differs
py-code2525match
py-indent1515match
json2020match
html1616match
regex4749differs
camel76differs
snake87differs
rare-word-x52631differs
repeat-tok4141match
base643431differs
hex1213differs
url1718differs
uuid2727match

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
FieldDeepSeek: DeepSeek V4 Pro 0813LiquidAI: LFM2.5-2.6B (free)Verdict
Context length104857665536× differs
Max output tokens8192× differs
Supported parametersfrequency_penalty, include_reasoning, logit_bias, logprobs, max_tokens, min_p, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, top_pfrequency_penalty, include_reasoning, logprobs, max_completion_tokens, 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":1,"top_p":1}{"temperature":0.1,"top_k":50,"repetition_penalty":1.1}× differs
Reasoning contract{"mandatory":false,"supported_efforts":["max","high","low"],"default_effort":"high"}{"mandatory":true}× differs
Modalitiestext->texttext->text✓ match

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.