AI Desk · Model Identity · Compare

OpenAI: GPT-5 Nano · Z.ai: GLM 5.3

Identity similarity: lowDivergent fingerprints · 18/50These models tokenise differently. Divergence on the high-information strings (CJK, emoji, digits) separates unrelated vocabularies within a handful of samples.

OpenAI: GPT-5 Nano recordZ.ai: GLM 5.3 recordOpen in the interactive bench

Tokenizer fingerprint

18 of 50 mutually clean strings agree
Test stringOpenAI: GPT-5 NanoZ.ai: GLM 5.3Verdict
en-prose1414match
en-long1716differs
spaces-2033match
spaces-6033match
tabs-2033match
newlines-2044match
mixed-ws65differs
digits-935differs
digits-1247differs
digits-301010match
digits-sep79differs
float-long914differs
zh-common129differs
zh-long1710differs
zh-rare2018differs
ja-kana1010match
ja-kanji1314differs
ko1117differs
ru1113differs
ar1113differs
he1024differs
hi1127differs
th1124differs
el1414match
emoji-basic85differs
emoji-skin1310differs
emoji-zwj-family117differs
emoji-zwj-x33321differs
emoji-flags1624differs
emoji-prof2015differs
math3127differs
boxdraw2728differs
combining1110differs
cjk-ext-b1313match
surrogates4040match
zalgo3522differs
rtl-mix58differs
py-code2525match
py-indent1515match
json2021differs
html1616match
regex4444match
camel65differs
snake66match
rare-word-x53131match
repeat-tok2222match
base643435differs
hex1114differs
url1717match
uuid2728differs

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
FieldOpenAI: GPT-5 NanoZ.ai: GLM 5.3Verdict
Context length4000001048576× differs
Max output tokens128000131072× differs
Supported parametersinclude_reasoning, max_completion_tokens, max_tokens, reasoning, reasoning_effort, response_format, seed, structured_outputs, tool_choice, toolsinclude_reasoning, max_tokens, reasoning, reasoning_effort, response_format, temperature, tool_choice, tools, top_k, 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}× differs
Reasoning contract{"mandatory":true,"supported_efforts":["high","medium","low","minimal"],"default_effort":"medium"}{"mandatory":true,"default_enabled":true,"supported_efforts":["max","high","low"],"default_effort":"max"}× differs
Modalitiestext+image+file->texttext->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.