{
 "version": "0.1",
 "generated": "2026-10-17",
 "method": {
  "corpus": "FLORES+ (openlanguagedata/flores_plus), dev + devtest, CC BY-SA 4.0",
  "sentences": 2009,
  "metric": "sum(Arabic tokens) / sum(English tokens) over identical parallel sentences",
  "ci": "95% percentile bootstrap over sentences, 2000 resamples, seed 42",
  "special_tokens": "excluded",
  "environment": {
   "date": "2026-10-17",
   "python": "3.9.7",
   "platform": "macOS-10.16-x86_64-i386-64bit",
   "transformers": "4.57.6",
   "tiktoken": "0.14.0"
  }
 },
 "varieties": [
  {
   "id": "eng",
   "flores": "eng_Latn",
   "label_ar": "الإنجليزية",
   "label_en": "English"
  },
  {
   "id": "msa",
   "flores": "arb_Arab",
   "label_ar": "الفصحى",
   "label_en": "Modern Standard Arabic"
  },
  {
   "id": "najdi",
   "flores": "ars_Arab",
   "label_ar": "النجدية",
   "label_en": "Najdi"
  },
  {
   "id": "mesopotamian",
   "flores": "acm_Arab",
   "label_ar": "العراقية",
   "label_en": "Mesopotamian"
  },
  {
   "id": "egyptian",
   "flores": "arz_Arab",
   "label_ar": "المصرية",
   "label_en": "Egyptian"
  },
  {
   "id": "levantine_n",
   "flores": "apc_Arab_nort3139",
   "label_ar": "الشامية الشمالية",
   "label_en": "North Levantine"
  },
  {
   "id": "levantine_s",
   "flores": "apc_Arab_sout3123",
   "label_ar": "الشامية الجنوبية",
   "label_en": "South Levantine"
  },
  {
   "id": "moroccan",
   "flores": "ary_Arab",
   "label_ar": "المغربية",
   "label_en": "Moroccan"
  }
 ],
 "tokenizers": [
  {
   "id": "o200k",
   "source": "o200k_base",
   "vocab": 200019,
   "revision": "tiktoken 0.14.0",
   "status": "measured",
   "ratios": {
    "eng": {
     "ratio": 1.0,
     "ci95": [
      1.0,
      1.0
     ],
     "tokens_per_word": 1.223
    },
    "msa": {
     "ratio": 1.382,
     "ci95": [
      1.372,
      1.392
     ],
     "tokens_per_word": 1.886
    },
    "najdi": {
     "ratio": 1.387,
     "ci95": [
      1.377,
      1.397
     ],
     "tokens_per_word": 1.889
    },
    "mesopotamian": {
     "ratio": 1.386,
     "ci95": [
      1.376,
      1.397
     ],
     "tokens_per_word": 1.982
    },
    "egyptian": {
     "ratio": 1.408,
     "ci95": [
      1.397,
      1.419
     ],
     "tokens_per_word": 1.932
    },
    "levantine_n": {
     "ratio": 1.351,
     "ci95": [
      1.342,
      1.361
     ],
     "tokens_per_word": 2.016
    },
    "levantine_s": {
     "ratio": 1.341,
     "ci95": [
      1.332,
      1.35
     ],
     "tokens_per_word": 1.904
    },
    "moroccan": {
     "ratio": 1.484,
     "ci95": [
      1.472,
      1.496
     ],
     "tokens_per_word": 2.078
    }
   },
   "name": "o200k (GPT-4o)",
   "provider": "OpenAI",
   "license": "proprietary (tokenizer public via tiktoken, MIT)",
   "hf_model": "Xenova/gpt-4o",
   "ratio": 1.382,
   "ci95": [
    1.372,
    1.392
   ],
   "arabic_tokens_per_word": 1.886,
   "english_tokens_per_word": 1.223
  },
  {
   "id": "qwen3",
   "source": "Qwen/Qwen3-4B",
   "vocab": 151669,
   "revision": "1cfa9a7208912126459214e8b04321603b3df60c",
   "status": "measured",
   "ratios": {
    "eng": {
     "ratio": 1.0,
     "ci95": [
      1.0,
      1.0
     ],
     "tokens_per_word": 1.257
    },
    "msa": {
     "ratio": 1.629,
     "ci95": [
      1.617,
      1.642
     ],
     "tokens_per_word": 2.286
    },
    "najdi": {
     "ratio": 1.633,
     "ci95": [
      1.621,
      1.647
     ],
     "tokens_per_word": 2.288
    },
    "mesopotamian": {
     "ratio": 1.622,
     "ci95": [
      1.61,
      1.635
     ],
     "tokens_per_word": 2.384
    },
    "egyptian": {
     "ratio": 1.636,
     "ci95": [
      1.622,
      1.649
     ],
     "tokens_per_word": 2.307
    },
    "levantine_n": {
     "ratio": 1.581,
     "ci95": [
      1.569,
      1.592
     ],
     "tokens_per_word": 2.424
    },
    "levantine_s": {
     "ratio": 1.55,
     "ci95": [
      1.539,
      1.561
     ],
     "tokens_per_word": 2.263
    },
    "moroccan": {
     "ratio": 1.695,
     "ci95": [
      1.68,
      1.709
     ],
     "tokens_per_word": 2.44
    }
   },
   "name": "Qwen3",
   "provider": "Alibaba",
   "license": "Apache 2.0",
   "hf_model": "Qwen/Qwen3-4B",
   "ratio": 1.629,
   "ci95": [
    1.617,
    1.642
   ],
   "arabic_tokens_per_word": 2.286,
   "english_tokens_per_word": 1.257
  },
  {
   "id": "llama3.2",
   "source": "meta-llama/Llama-3.2-1B-Instruct",
   "vocab": 128256,
   "revision": "9213176726f574b556790deb65791e0c5aa438b6",
   "status": "measured",
   "ratios": {
    "eng": {
     "ratio": 1.0,
     "ci95": [
      1.0,
      1.0
     ],
     "tokens_per_word": 1.236
    },
    "msa": {
     "ratio": 1.656,
     "ci95": [
      1.644,
      1.668
     ],
     "tokens_per_word": 2.284
    },
    "najdi": {
     "ratio": 1.663,
     "ci95": [
      1.651,
      1.675
     ],
     "tokens_per_word": 2.29
    },
    "mesopotamian": {
     "ratio": 1.649,
     "ci95": [
      1.637,
      1.661
     ],
     "tokens_per_word": 2.382
    },
    "egyptian": {
     "ratio": 1.649,
     "ci95": [
      1.636,
      1.662
     ],
     "tokens_per_word": 2.287
    },
    "levantine_n": {
     "ratio": 1.593,
     "ci95": [
      1.582,
      1.605
     ],
     "tokens_per_word": 2.402
    },
    "levantine_s": {
     "ratio": 1.576,
     "ci95": [
      1.565,
      1.587
     ],
     "tokens_per_word": 2.261
    },
    "moroccan": {
     "ratio": 1.706,
     "ci95": [
      1.692,
      1.72
     ],
     "tokens_per_word": 2.415
    }
   },
   "name": "Llama 3.2",
   "provider": "Meta",
   "license": "Llama 3.2 Community License",
   "hf_model": "Xenova/llama-3-tokenizer",
   "hf_model_note": "Llama 3.2 shares the Llama 3 tokenizer (128,256 tokens); the official repo is gated, so the browser loads a public mirror.",
   "ratio": 1.656,
   "ci95": [
    1.644,
    1.668
   ],
   "arabic_tokens_per_word": 2.284,
   "english_tokens_per_word": 1.236
  },
  {
   "id": "gemma3",
   "source": "google/gemma-3-1b-it",
   "vocab": 262145,
   "revision": "dcc83ea841ab6100d6b47a070329e1ba4cf78752",
   "status": "measured",
   "ratios": {
    "eng": {
     "ratio": 1.0,
     "ci95": [
      1.0,
      1.0
     ],
     "tokens_per_word": 1.237
    },
    "msa": {
     "ratio": 1.47,
     "ci95": [
      1.459,
      1.481
     ],
     "tokens_per_word": 2.028
    },
    "najdi": {
     "ratio": 1.474,
     "ci95": [
      1.463,
      1.486
     ],
     "tokens_per_word": 2.031
    },
    "mesopotamian": {
     "ratio": 1.461,
     "ci95": [
      1.45,
      1.472
     ],
     "tokens_per_word": 2.112
    },
    "egyptian": {
     "ratio": 1.436,
     "ci95": [
      1.424,
      1.447
     ],
     "tokens_per_word": 1.992
    },
    "levantine_n": {
     "ratio": 1.415,
     "ci95": [
      1.404,
      1.425
     ],
     "tokens_per_word": 2.135
    },
    "levantine_s": {
     "ratio": 1.389,
     "ci95": [
      1.379,
      1.399
     ],
     "tokens_per_word": 1.994
    },
    "moroccan": {
     "ratio": 1.519,
     "ci95": [
      1.507,
      1.531
     ],
     "tokens_per_word": 2.151
    }
   },
   "name": "Gemma 3",
   "provider": "Google",
   "license": "Gemma Terms of Use",
   "hf_model": null,
   "hf_model_note": "gated repo; measured locally, not loadable in the browser",
   "ratio": 1.47,
   "ci95": [
    1.459,
    1.481
   ],
   "arabic_tokens_per_word": 2.028,
   "english_tokens_per_word": 1.237
  },
  {
   "id": "mistral-tekken",
   "source": "mistralai/Mistral-Nemo-Instruct-2407",
   "vocab": 131072,
   "revision": "04d8a90549d23fc6bd7f642064003592df51e9b3",
   "status": "measured",
   "ratios": {
    "eng": {
     "ratio": 1.0,
     "ci95": [
      1.0,
      1.0
     ],
     "tokens_per_word": 1.267
    },
    "msa": {
     "ratio": 1.245,
     "ci95": [
      1.236,
      1.254
     ],
     "tokens_per_word": 1.761
    },
    "najdi": {
     "ratio": 1.25,
     "ci95": [
      1.241,
      1.26
     ],
     "tokens_per_word": 1.765
    },
    "mesopotamian": {
     "ratio": 1.262,
     "ci95": [
      1.253,
      1.271
     ],
     "tokens_per_word": 1.869
    },
    "egyptian": {
     "ratio": 1.288,
     "ci95": [
      1.278,
      1.297
     ],
     "tokens_per_word": 1.831
    },
    "levantine_n": {
     "ratio": 1.236,
     "ci95": [
      1.227,
      1.244
     ],
     "tokens_per_word": 1.911
    },
    "levantine_s": {
     "ratio": 1.229,
     "ci95": [
      1.221,
      1.237
     ],
     "tokens_per_word": 1.808
    },
    "moroccan": {
     "ratio": 1.37,
     "ci95": [
      1.359,
      1.382
     ],
     "tokens_per_word": 1.988
    }
   },
   "name": "Mistral Tekken",
   "provider": "Mistral AI",
   "license": "Apache 2.0",
   "hf_model": "mistralai/Mistral-Nemo-Instruct-2407",
   "ratio": 1.245,
   "ci95": [
    1.236,
    1.254
   ],
   "arabic_tokens_per_word": 1.761,
   "english_tokens_per_word": 1.267
  },
  {
   "id": "allam",
   "source": "humain-ai/ALLaM-7B-Instruct-preview",
   "vocab": 64000,
   "revision": "a28dd1e67420cde72d3629c8633a974cf7d9c366",
   "status": "measured",
   "ratios": {
    "eng": {
     "ratio": 1.0,
     "ci95": [
      1.0,
      1.0
     ],
     "tokens_per_word": 1.339
    },
    "msa": {
     "ratio": 0.966,
     "ci95": [
      0.959,
      0.973
     ],
     "tokens_per_word": 1.444
    },
    "najdi": {
     "ratio": 0.969,
     "ci95": [
      0.962,
      0.976
     ],
     "tokens_per_word": 1.446
    },
    "mesopotamian": {
     "ratio": 0.958,
     "ci95": [
      0.951,
      0.964
     ],
     "tokens_per_word": 1.499
    },
    "egyptian": {
     "ratio": 0.973,
     "ci95": [
      0.965,
      0.98
     ],
     "tokens_per_word": 1.462
    },
    "levantine_n": {
     "ratio": 0.942,
     "ci95": [
      0.936,
      0.949
     ],
     "tokens_per_word": 1.54
    },
    "levantine_s": {
     "ratio": 0.936,
     "ci95": [
      0.93,
      0.942
     ],
     "tokens_per_word": 1.456
    },
    "moroccan": {
     "ratio": 1.075,
     "ci95": [
      1.066,
      1.084
     ],
     "tokens_per_word": 1.648
    }
   },
   "name": "ALLaM",
   "provider": "HUMAIN / SDAIA",
   "license": "ALLaM license (gated)",
   "hf_model": null,
   "hf_model_note": "gated repo; measured locally, not loadable in the browser",
   "ratio": 0.966,
   "ci95": [
    0.959,
    0.973
   ],
   "arabic_tokens_per_word": 1.444,
   "english_tokens_per_word": 1.339
  },
  {
   "id": "jais",
   "source": "inceptionai/jais-family-6p7b-chat",
   "vocab": 84992,
   "revision": "683805efe6126c6536feb4aa23317e70222ac94c",
   "status": "measured",
   "ratios": {
    "eng": {
     "ratio": 1.0,
     "ci95": [
      1.0,
      1.0
     ],
     "tokens_per_word": 1.268
    },
    "msa": {
     "ratio": 1.013,
     "ci95": [
      1.005,
      1.021
     ],
     "tokens_per_word": 1.433
    },
    "najdi": {
     "ratio": 1.016,
     "ci95": [
      1.008,
      1.025
     ],
     "tokens_per_word": 1.436
    },
    "mesopotamian": {
     "ratio": 0.982,
     "ci95": [
      0.975,
      0.99
     ],
     "tokens_per_word": 1.455
    },
    "egyptian": {
     "ratio": 1.004,
     "ci95": [
      0.996,
      1.012
     ],
     "tokens_per_word": 1.428
    },
    "levantine_n": {
     "ratio": 0.972,
     "ci95": [
      0.965,
      0.979
     ],
     "tokens_per_word": 1.503
    },
    "levantine_s": {
     "ratio": 0.957,
     "ci95": [
      0.951,
      0.963
     ],
     "tokens_per_word": 1.408
    },
    "moroccan": {
     "ratio": 1.103,
     "ci95": [
      1.092,
      1.113
     ],
     "tokens_per_word": 1.6
    }
   },
   "name": "Jais (family 6.7B)",
   "provider": "Inception / G42",
   "license": "Apache 2.0 (gated repo)",
   "hf_model": "derek-thomas/jais-13b-chat-hf",
   "hf_model_note": "official repo is gated; the browser loads a community mirror with the same 84,992-token vocabulary",
   "ratio": 1.013,
   "ci95": [
    1.005,
    1.021
   ],
   "arabic_tokens_per_word": 1.433,
   "english_tokens_per_word": 1.268
  },
  {
   "id": "jais2",
   "source": "inception42/Jais-2-8B-Chat",
   "vocab": 150222,
   "revision": "da0e1639cd92b508b24120f7f77f5270a8465dc4",
   "status": "measured",
   "ratios": {
    "eng": {
     "ratio": 1.0,
     "ci95": [
      1.0,
      1.0
     ],
     "tokens_per_word": 1.249
    },
    "msa": {
     "ratio": 1.047,
     "ci95": [
      1.038,
      1.057
     ],
     "tokens_per_word": 1.46
    },
    "najdi": {
     "ratio": 1.05,
     "ci95": [
      1.041,
      1.059
     ],
     "tokens_per_word": 1.461
    },
    "mesopotamian": {
     "ratio": 1.001,
     "ci95": [
      0.993,
      1.008
     ],
     "tokens_per_word": 1.461
    },
    "egyptian": {
     "ratio": 1.018,
     "ci95": [
      1.01,
      1.026
     ],
     "tokens_per_word": 1.427
    },
    "levantine_n": {
     "ratio": 1.0,
     "ci95": [
      0.993,
      1.008
     ],
     "tokens_per_word": 1.524
    },
    "levantine_s": {
     "ratio": 0.995,
     "ci95": [
      0.988,
      1.001
     ],
     "tokens_per_word": 1.442
    },
    "moroccan": {
     "ratio": 1.128,
     "ci95": [
      1.117,
      1.139
     ],
     "tokens_per_word": 1.613
    }
   },
   "name": "Jais 2",
   "provider": "Inception",
   "license": "gated repo",
   "hf_model": null,
   "hf_model_note": "gated repo; measured locally, not loadable in the browser",
   "ratio": 1.047,
   "ci95": [
    1.038,
    1.057
   ],
   "arabic_tokens_per_word": 1.46,
   "english_tokens_per_word": 1.249
  },
  {
   "id": "jais3",
   "source": "IFM/Jais-3-7B-FP8",
   "vocab": 250624,
   "revision": "10d6507ddb2f43e34f9c4d3ae37c89f61f2685de",
   "status": "measured",
   "ratios": {
    "eng": {
     "ratio": 1.0,
     "ci95": [
      1.0,
      1.0
     ],
     "tokens_per_word": 1.21
    },
    "msa": {
     "ratio": 1.05,
     "ci95": [
      1.042,
      1.058
     ],
     "tokens_per_word": 1.418
    },
    "najdi": {
     "ratio": 1.054,
     "ci95": [
      1.046,
      1.062
     ],
     "tokens_per_word": 1.421
    },
    "mesopotamian": {
     "ratio": 1.033,
     "ci95": [
      1.025,
      1.04
     ],
     "tokens_per_word": 1.461
    },
    "egyptian": {
     "ratio": 1.048,
     "ci95": [
      1.041,
      1.056
     ],
     "tokens_per_word": 1.424
    },
    "levantine_n": {
     "ratio": 1.012,
     "ci95": [
      1.005,
      1.019
     ],
     "tokens_per_word": 1.494
    },
    "levantine_s": {
     "ratio": 1.006,
     "ci95": [
      0.999,
      1.012
     ],
     "tokens_per_word": 1.413
    },
    "moroccan": {
     "ratio": 1.14,
     "ci95": [
      1.131,
      1.15
     ],
     "tokens_per_word": 1.58
    }
   },
   "name": "Jais 3 (3.7B–375B)",
   "provider": "MBZUAI IFM / Inception",
   "license": "Apache 2.0",
   "hf_model": "IFM/Jais-3-7B-FP8",
   "ratio": 1.05,
   "ci95": [
    1.042,
    1.058
   ],
   "arabic_tokens_per_word": 1.418,
   "english_tokens_per_word": 1.21
  },
  {
   "id": "jais3-0.9b",
   "source": "IFM/Jais-3-0.9B-MLX-bf16",
   "vocab": 64256,
   "revision": "1d5cf547ae5292eaf066408bcf763b4aa21c03e9",
   "status": "measured",
   "ratios": {
    "eng": {
     "ratio": 1.0,
     "ci95": [
      1.0,
      1.0
     ],
     "tokens_per_word": 1.31
    },
    "msa": {
     "ratio": 1.165,
     "ci95": [
      1.156,
      1.174
     ],
     "tokens_per_word": 1.703
    },
    "najdi": {
     "ratio": 1.168,
     "ci95": [
      1.16,
      1.177
     ],
     "tokens_per_word": 1.705
    },
    "mesopotamian": {
     "ratio": 1.163,
     "ci95": [
      1.154,
      1.171
     ],
     "tokens_per_word": 1.78
    },
    "egyptian": {
     "ratio": 1.176,
     "ci95": [
      1.167,
      1.185
     ],
     "tokens_per_word": 1.728
    },
    "levantine_n": {
     "ratio": 1.143,
     "ci95": [
      1.135,
      1.151
     ],
     "tokens_per_word": 1.826
    },
    "levantine_s": {
     "ratio": 1.127,
     "ci95": [
      1.12,
      1.134
     ],
     "tokens_per_word": 1.713
    },
    "moroccan": {
     "ratio": 1.264,
     "ci95": [
      1.253,
      1.275
     ],
     "tokens_per_word": 1.896
    }
   },
   "name": "Jais 3 0.9B",
   "provider": "MBZUAI IFM / Inception",
   "license": "Apache 2.0",
   "hf_model": "IFM/Jais-3-0.9B-MLX-bf16",
   "ratio": 1.165,
   "ci95": [
    1.156,
    1.174
   ],
   "arabic_tokens_per_word": 1.703,
   "english_tokens_per_word": 1.31
  },
  {
   "id": "fanar",
   "source": "QCRI/Fanar-1-9B-Instruct",
   "vocab": 128256,
   "revision": "d30938d0efb1fd251727ee96df6987802ff84662",
   "status": "measured",
   "ratios": {
    "eng": {
     "ratio": 1.0,
     "ci95": [
      1.0,
      1.0
     ],
     "tokens_per_word": 1.269
    },
    "msa": {
     "ratio": 1.453,
     "ci95": [
      1.442,
      1.465
     ],
     "tokens_per_word": 2.058
    },
    "najdi": {
     "ratio": 1.459,
     "ci95": [
      1.448,
      1.47
     ],
     "tokens_per_word": 2.062
    },
    "mesopotamian": {
     "ratio": 1.458,
     "ci95": [
      1.447,
      1.469
     ],
     "tokens_per_word": 2.163
    },
    "egyptian": {
     "ratio": 1.458,
     "ci95": [
      1.446,
      1.469
     ],
     "tokens_per_word": 2.076
    },
    "levantine_n": {
     "ratio": 1.412,
     "ci95": [
      1.402,
      1.423
     ],
     "tokens_per_word": 2.187
    },
    "levantine_s": {
     "ratio": 1.391,
     "ci95": [
      1.381,
      1.401
     ],
     "tokens_per_word": 2.05
    },
    "moroccan": {
     "ratio": 1.522,
     "ci95": [
      1.509,
      1.535
     ],
     "tokens_per_word": 2.212
    }
   },
   "name": "Fanar",
   "provider": "QCRI",
   "license": "gated repo",
   "hf_model": null,
   "hf_model_note": "gated repo; measured locally, not loadable in the browser",
   "ratio": 1.453,
   "ci95": [
    1.442,
    1.465
   ],
   "arabic_tokens_per_word": 2.058,
   "english_tokens_per_word": 1.269
  }
 ],
 "custom_sets": {
  "gulf_v0": {
   "source": "drafted with AI assistance (Claude), reviewed and edited by a native Gulf speaker; corpus/gulf_v0.txt, MIT",
   "items": 20,
   "results": {
    "o200k": {
     "gulf_vs_en": 1.149,
     "gulf_vs_en_ci95": "1.071–1.222",
     "msa_vs_en": 1.213,
     "msa_vs_en_ci95": "1.120–1.313",
     "gulf_vs_msa": 0.947,
     "gulf_vs_msa_ci95": "0.889–1.002"
    },
    "qwen3": {
     "gulf_vs_en": 1.243,
     "gulf_vs_en_ci95": "1.176–1.310",
     "msa_vs_en": 1.308,
     "msa_vs_en_ci95": "1.215–1.405",
     "gulf_vs_msa": 0.951,
     "gulf_vs_msa_ci95": "0.893–1.011"
    },
    "llama3.2": {
     "gulf_vs_en": 1.24,
     "gulf_vs_en_ci95": "1.165–1.307",
     "msa_vs_en": 1.39,
     "msa_vs_en_ci95": "1.296–1.489",
     "gulf_vs_msa": 0.892,
     "gulf_vs_msa_ci95": "0.837–0.943"
    },
    "gemma3": {
     "gulf_vs_en": 1.043,
     "gulf_vs_en_ci95": "0.982–1.102",
     "msa_vs_en": 1.193,
     "msa_vs_en_ci95": "1.107–1.282",
     "gulf_vs_msa": 0.874,
     "gulf_vs_msa_ci95": "0.810–0.934"
    },
    "mistral-tekken": {
     "gulf_vs_en": 1.049,
     "gulf_vs_en_ci95": "0.978–1.109",
     "msa_vs_en": 1.07,
     "msa_vs_en_ci95": "1.005–1.139",
     "gulf_vs_msa": 0.981,
     "gulf_vs_msa_ci95": "0.916–1.040"
    },
    "allam": {
     "gulf_vs_en": 0.729,
     "gulf_vs_en_ci95": "0.663–0.789",
     "msa_vs_en": 0.77,
     "msa_vs_en_ci95": "0.710–0.827",
     "gulf_vs_msa": 0.947,
     "gulf_vs_msa_ci95": "0.891–1.000"
    },
    "jais": {
     "gulf_vs_en": 0.81,
     "gulf_vs_en_ci95": "0.752–0.864",
     "msa_vs_en": 0.894,
     "msa_vs_en_ci95": "0.824–0.952",
     "gulf_vs_msa": 0.907,
     "gulf_vs_msa_ci95": "0.842–0.981"
    },
    "jais2": {
     "gulf_vs_en": 0.819,
     "gulf_vs_en_ci95": "0.764–0.870",
     "msa_vs_en": 0.907,
     "msa_vs_en_ci95": "0.835–0.969",
     "gulf_vs_msa": 0.904,
     "gulf_vs_msa_ci95": "0.840–0.978"
    },
    "jais3": {
     "gulf_vs_en": 0.838,
     "gulf_vs_en_ci95": "0.782–0.890",
     "msa_vs_en": 0.874,
     "msa_vs_en_ci95": "0.807–0.937",
     "gulf_vs_msa": 0.96,
     "gulf_vs_msa_ci95": "0.904–1.026"
    },
    "jais3-0.9b": {
     "gulf_vs_en": 0.932,
     "gulf_vs_en_ci95": "0.876–0.990",
     "msa_vs_en": 0.994,
     "msa_vs_en_ci95": "0.939–1.052",
     "gulf_vs_msa": 0.937,
     "gulf_vs_msa_ci95": "0.880–0.991"
    },
    "fanar": {
     "gulf_vs_en": 1.099,
     "gulf_vs_en_ci95": "1.022–1.167",
     "msa_vs_en": 1.226,
     "msa_vs_en_ci95": "1.147–1.307",
     "gulf_vs_msa": 0.896,
     "gulf_vs_msa_ci95": "0.847–0.947"
    }
   }
  }
 },
 "updated": "2026-10-17",
 "default_tokenizer": "o200k",
 "note": "ratio = مجموع رموز النص العربي ÷ مجموع رموز النص الإنجليزي على الجمل المتوازية نفسها (FLORES+). الحقول العلوية ratio/ci95/tokens_per_word هي صف الفصحى (msa) منسوخًا لتوافق الصفحات؛ التفاصيل لكل لهجة في ratios."
}