{
 "product": "YFarmX Model Identity",
 "method": "tokenizer fingerprinting v1-50",
 "updated": "2026-08-22",
 "record_url": "https://yfarmx.com/ai/models/microsoft/phi-4/",
 "citation": "YFarmX Model Identity: Microsoft: Phi 4 (microsoft/phi-4) evidence record. Measured 2026-08-21. https://yfarmx.com/ai/models/microsoft/phi-4/",
 "record": {
  "id": "microsoft/phi-4",
  "slug": "microsoft/phi-4",
  "name": "Microsoft: Phi 4",
  "lab": "Microsoft",
  "alias": false,
  "variant_of": null,
  "created": 1736489872,
  "new_since_20_aug": false,
  "description": "[Microsoft Research](/microsoft) Phi-4 is designed to perform well in complex reasoning tasks and can operate efficiently in situations with limited memory or where quick responses are needed. At 14 billion...",
  "declared": {
   "tokenizer": "Other",
   "modality": "text->text",
   "input": [
    "text"
   ],
   "context": 16384,
   "max_output": 16384,
   "pricing": {
    "prompt": "0.00000007",
    "completion": "0.00000014"
   },
   "params": [
    "frequency_penalty",
    "logit_bias",
    "max_tokens",
    "min_p",
    "presence_penalty",
    "repetition_penalty",
    "response_format",
    "seed",
    "stop",
    "structured_outputs",
    "temperature",
    "top_k",
    "top_p"
   ],
   "defaults": {},
   "reasoning": null,
   "hugging_face": "microsoft/phi-4"
  },
  "measurement": {
   "status": "measured",
   "at": "2026-08-21",
   "run": "catalogue-sweep",
   "overhead": 7,
   "providers": [
    "DeepInfra"
   ],
   "clean_rows": 50,
   "corrupt_rows": 0,
   "signature": "tk_ba918336",
   "counts": {
    "en-prose": 14,
    "en-long": 16,
    "spaces-20": 3,
    "spaces-60": 3,
    "tabs-20": 3,
    "newlines-20": 4,
    "mixed-ws": 5,
    "digits-9": 3,
    "digits-12": 4,
    "digits-30": 10,
    "digits-sep": 7,
    "float-long": 9,
    "zh-common": 18,
    "zh-long": 25,
    "zh-rare": 23,
    "ja-kana": 14,
    "ja-kanji": 17,
    "ko": 19,
    "ru": 16,
    "ar": 25,
    "he": 24,
    "hi": 32,
    "th": 25,
    "el": 29,
    "emoji-basic": 10,
    "emoji-skin": 30,
    "emoji-zwj-family": 18,
    "emoji-zwj-x3": 54,
    "emoji-flags": 24,
    "emoji-prof": 29,
    "math": 32,
    "boxdraw": 28,
    "combining": 14,
    "cjk-ext-b": 13,
    "surrogates": 40,
    "zalgo": 36,
    "rtl-mix": 11,
    "py-code": 25,
    "py-indent": 15,
    "json": 20,
    "html": 16,
    "regex": 44,
    "camel": 5,
    "snake": 6,
    "rare-word-x5": 31,
    "repeat-tok": 22,
    "base64": 35,
    "hex": 11,
    "url": 17,
    "uuid": 27
   },
   "corrupt": {},
   "observations": [
    {
     "at": "2026-08-21",
     "run": "catalogue-sweep",
     "clean": 50
    }
   ]
  },
  "note": {
   "date": "2026-08-21",
   "text": "Phi-4's repository tokenizer, run over the 50 strings offline, matches OpenAI's open cl100k_base encoding on every string (vocabulary 100,352, the padded cl100k table). IBM's Granite 4.0 and 4.1 declare the same vocabulary size and their endpoints return identical counts, which is why Microsoft and IBM share an exact signature group: both build on the tokenizer OpenAI released as open source with tiktoken.",
   "sources": [
    {
     "label": "Phi-4 repository (config.json: vocab_size 100352)",
     "url": "https://huggingface.co/microsoft/phi-4/blob/main/config.json"
    },
    {
     "label": "Granite 4.0 H Micro repository (config.json: vocab_size 100352)",
     "url": "https://huggingface.co/ibm-granite/granite-4.0-h-micro/blob/main/config.json"
    },
    {
     "label": "OpenAI tiktoken (cl100k_base, open source)",
     "url": "https://github.com/openai/tiktoken"
    }
   ]
  },
  "api_signature": "api_7f4ca365",
  "api_matches": [],
  "identity": {
   "kind": "lineage",
   "family": "OpenAI cl100k",
   "confidence": "high",
   "confidence_reason": "an exact fingerprint shared with 2 models",
   "basis": [
    "exact tokenizer signature shared with 2 models"
   ],
   "contradiction": null,
   "declared_vs_measured": "resolved"
  },
  "comparator": "ibm-granite/granite-4.1-8b",
  "neighbours": [
   {
    "id": "ibm-granite/granite-4.0-h-micro",
    "same": 50,
    "total": 50
   },
   {
    "id": "ibm-granite/granite-4.1-8b",
    "same": 50,
    "total": 50
   },
   {
    "id": "meta-llama/llama-3.2-1b-instruct",
    "same": 31,
    "total": 50
   },
   {
    "id": "nousresearch/hermes-4-70b",
    "same": 31,
    "total": 50
   },
   {
    "id": "nousresearch/hermes-3-llama-3.1-70b",
    "same": 31,
    "total": 50
   },
   {
    "id": "nousresearch/hermes-3-llama-3.1-405b",
    "same": 31,
    "total": 50
   }
  ]
 }
}