{
 "tabs": [
  {
   "id": "overview",
   "title": "Overview",
   "sort": "arena",
   "blurb": "Our checkpoints on every benchmark, one headline number each (WER %, lower is better). The other tabs rank them next to other models where those were run.",
   "links": [],
   "cols": [
    {
     "k": "arena",
     "l": "Arabic Arena",
     "d": 2,
     "higher": false
    },
    {
     "k": "lbsets",
     "l": "Leaderboard sets",
     "d": 2,
     "higher": false
    },
    {
     "k": "kejue",
     "l": "kejue",
     "d": 2,
     "higher": false
    },
    {
     "k": "codeswitch",
     "l": "Code-switch",
     "d": 2,
     "higher": false
    },
    {
     "k": "testarabic",
     "l": "test-arabic",
     "d": 2,
     "higher": false
    },
    {
     "k": "smartturn",
     "l": "smart-turn",
     "d": 2,
     "higher": false
    },
    {
     "k": "english",
     "l": "English Arena",
     "d": 2,
     "higher": false
    }
   ],
   "rows": [
    {
     "model": "qwen3-asr-1.7b-ar-run5",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5",
     "kind": "gated",
     "ours": "run5",
     "v": {
      "arena": 16.408964013353454,
      "lbsets": 17.82,
      "kejue": 19.08082540549202,
      "codeswitch": 22.010892333472977,
      "testarabic": 11.963787122429984,
      "smartturn": 4.8642455206,
      "english": 10.787542312798355
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5base",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5base",
     "kind": "gated",
     "ours": "run5base",
     "v": {
      "arena": 18.635375913757237,
      "lbsets": 18.26,
      "kejue": 19.2775624781156,
      "codeswitch": 22.320904901550062,
      "testarabic": 11.786107115661224,
      "smartturn": 4.7134501718,
      "english": 10.240182650032354
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5-comboB",
     "url": null,
     "kind": "private",
     "ours": "comboB",
     "v": {
      "arena": null,
      "lbsets": 18.14,
      "kejue": 18.83156576596894,
      "codeswitch": 22.547130289065773,
      "testarabic": 11.777646162957948,
      "smartturn": null,
      "english": 10.793840727720923
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run4",
     "url": null,
     "kind": "private",
     "ours": "run4",
     "v": {
      "arena": null,
      "lbsets": 17.92,
      "kejue": 19.212873666906034,
      "codeswitch": 28.051948051948052,
      "testarabic": 11.980709027836536,
      "smartturn": null,
      "english": 11.12928336079596
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run3",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run3",
     "kind": "gated",
     "ours": "run3",
     "v": {
      "arena": 18.670701813314007,
      "lbsets": 18.82,
      "kejue": 18.85500804159075,
      "codeswitch": 22.23711772098869,
      "testarabic": 16.49039681868178,
      "smartturn": 5.3679277626,
      "english": 19.719087125956708
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run2",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run2",
     "kind": "gated",
     "ours": "run2",
     "v": {
      "arena": null,
      "lbsets": 23.04,
      "kejue": null,
      "codeswitch": null,
      "testarabic": 16.355021575429394,
      "smartturn": 5.1890355026,
      "english": null
     }
    },
    {
     "model": "Qwen3-ASR-1.7B",
     "url": "https://huggingface.co/Qwen/Qwen3-ASR-1.7B",
     "kind": "open",
     "ours": "base",
     "v": {
      "arena": 29.964125579375036,
      "lbsets": 31.88,
      "kejue": null,
      "codeswitch": 46.63594470046083,
      "testarabic": 19.079448345883744,
      "smartturn": 6.5403938207,
      "english": 9.641583047808838
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6",
     "url": null,
     "kind": "private",
     "ours": "run6",
     "v": {
      "arena": 16.745246732294017,
      "lbsets": 18.03,
      "kejue": 19.590323978183847,
      "codeswitch": 22.3963133640553,
      "testarabic": 11.68457568322193,
      "smartturn": 4.6890909232,
      "english": 9.324433279704724
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6-soup50",
     "url": null,
     "kind": "private",
     "ours": "soup-r6-50",
     "v": {
      "arena": 16.41738102422492,
      "lbsets": 17.79,
      "kejue": 19.251152825832794,
      "codeswitch": 22.153330540427316,
      "testarabic": 11.14307471021237,
      "smartturn": 4.6140799035,
      "english": 9.519266746178067
     }
    }
   ]
  },
  {
   "id": "arena",
   "title": "Arabic Arena",
   "sort": "avg",
   "blurb": "The Arabic Arena's 15 test sets that are re-hosted on the Hub, run with the arena's own runner (no repetition repair). Clean, added noise, code-switching (Mixat), telephony codecs and call-center audio; the average weights the tracks Clean 35 %, Noisy 15 %, Code-switch 10 %, Telephony 20 %, Call-center 20 %. WER (%), lower is better.",
   "links": [
    [
     "Arabic Arena (GitHub)",
     "https://github.com/kadirnar/arabic-arena"
    ],
    [
     "Arena Space",
     "https://huggingface.co/spaces/kadirnar/arabic-arena"
    ]
   ],
   "cols": [
    {
     "k": "avg",
     "l": "Average",
     "d": 2,
     "higher": false
    },
    {
     "k": "clean",
     "l": "Clean",
     "d": 2,
     "higher": false
    },
    {
     "k": "noisy",
     "l": "Noisy",
     "d": 2,
     "higher": false
    },
    {
     "k": "codeswitch",
     "l": "Code-switch",
     "d": 2,
     "higher": false
    },
    {
     "k": "telephony",
     "l": "Telephony",
     "d": 2,
     "higher": false
    },
    {
     "k": "callcenter",
     "l": "Call-center",
     "d": 2,
     "higher": false
    }
   ],
   "rows": [
    {
     "model": "cohere-transcribe-arabic-07-2026",
     "url": "https://huggingface.co/CohereLabs/cohere-transcribe-arabic-07-2026",
     "kind": "open",
     "v": {
      "clean": 12.390658275615479,
      "noisy": 20.85374600845231,
      "codeswitch": 17.731329262564586,
      "telephony": 18.284332759938206,
      "callcenter": 35.77775312141849,
      "avg": 20.050342400261062
     }
    },
    {
     "model": "audar-asr-v1-turbo",
     "url": "https://huggingface.co/audarai/Audar-ASR-V1-Turbo",
     "kind": "open",
     "v": {
      "clean": 11.73107668207697,
      "noisy": 22.609985431598545,
      "codeswitch": 15.331924221074056,
      "telephony": 20.06084769601615,
      "callcenter": 44.310750516951536,
      "avg": 21.904886718167667
     }
    },
    {
     "model": "voxtral-mini-4b-realtime-arabic",
     "url": "https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-Arabic",
     "kind": "open",
     "v": {
      "clean": 15.425621855366309,
      "noisy": 27.285474661362546,
      "codeswitch": 17.609989040237984,
      "telephony": 27.245929339140783,
      "callcenter": 54.19439872640689,
      "avg": 27.540853365715922
     }
    },
    {
     "model": "lemura-arabic-asr-qwen3",
     "url": "https://huggingface.co/lemuralabs/lemura-arabic-asr-qwen3",
     "kind": "open",
     "v": {
      "clean": 16.13613032123206,
      "noisy": 26.68697717260017,
      "codeswitch": 23.39909190543291,
      "telephony": 24.7674972651552,
      "callcenter": 52.813325355575124,
      "avg": 27.506765903010603
     }
    },
    {
     "model": "nvidia-fastconformer-ar-pc",
     "url": "https://huggingface.co/nvidia/stt_ar_fastconformer_hybrid_large_pc_v1.0",
     "kind": "open",
     "v": {
      "clean": 15.016100363687137,
      "noisy": 24.11714446730387,
      "codeswitch": 27.853452325035228,
      "telephony": 23.72749645362849,
      "callcenter": 42.465952728083316,
      "avg": 24.897241866231962
     }
    },
    {
     "model": "lemura-arabic-asr-lite",
     "url": "https://huggingface.co/lemuralabs/lemura-arabic-asr-lite",
     "kind": "open",
     "v": {
      "clean": 15.492612394294596,
      "noisy": 27.46070524593256,
      "codeswitch": 43.7646782527008,
      "telephony": 24.348292899320384,
      "callcenter": 47.42399557406364,
      "avg": 28.272445644839877
     }
    },
    {
     "model": "cohere-transcribe-03-2026",
     "url": "https://huggingface.co/CohereLabs/cohere-transcribe-03-2026",
     "kind": "open",
     "v": {
      "clean": 17.531422335257375,
      "noisy": 28.39906409797558,
      "codeswitch": 49.45201189917019,
      "telephony": 29.685234232451695,
      "callcenter": 47.49571164650957,
      "avg": 30.77724779774569
     }
    },
    {
     "model": "Qwen3-ASR-1.7B",
     "url": "https://huggingface.co/Qwen/Qwen3-ASR-1.7B",
     "kind": "open",
     "ours": "base",
     "v": {
      "clean": 19.310350595386666,
      "noisy": 30.396897410864064,
      "codeswitch": 20.99577266322217,
      "telephony": 29.375462054604792,
      "callcenter": 53.356492910584585,
      "avg": 29.964125579375036
     }
    },
    {
     "model": "audar-asr-v1-flash",
     "url": "https://huggingface.co/audarai/Audar-ASR-V1-Flash",
     "kind": "open",
     "v": {
      "clean": 20.675662297397093,
      "noisy": 34.11812156033917,
      "codeswitch": 21.978237043995616,
      "telephony": 32.06737321422339,
      "callcenter": 57.4858993091238,
      "avg": 32.462678247208856
     }
    },
    {
     "model": "voxtral-mini-4b-realtime-2602",
     "url": "https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602",
     "kind": "open",
     "v": {
      "clean": 21.42176068849527,
      "noisy": 34.81248035046601,
      "codeswitch": 23.966651009863785,
      "telephony": 37.80990010458637,
      "callcenter": 66.36616730319376,
      "avg": 35.95136687608565
     }
    },
    {
     "model": "qwen3-asr-arabic-ksa",
     "url": "https://huggingface.co/vadimbelsky/qwen3-asr-arabic-ksa",
     "kind": "open",
     "v": {
      "clean": 23.667308405340375,
      "noisy": 37.145850638764735,
      "codeswitch": 23.01158603413183,
      "telephony": 34.78249002631434,
      "callcenter": 63.39572426438744,
      "avg": 35.79223699923738
     }
    },
    {
     "model": "nemotron-3.5-asr-arabic-dialectal-v2",
     "url": "https://huggingface.co/oddadmix/nemotron-3.5-asr-arabic-dialectal-v2",
     "kind": "open",
     "v": {
      "clean": 26.68971180611614,
      "noisy": 38.93618771612041,
      "codeswitch": 27.873023328636293,
      "telephony": 39.95512911816315,
      "callcenter": 64.08090897838203,
      "avg": 38.776337241731376
     }
    },
    {
     "model": "gemma-4-e2b-it",
     "url": "https://huggingface.co/google/gemma-4-E2B-it",
     "kind": "open",
     "v": {
      "clean": 24.71731169570815,
      "noisy": 40.964566319063124,
      "codeswitch": 27.64599968686394,
      "telephony": 38.24054930171625,
      "callcenter": 81.38613550975352,
      "avg": 41.48568097233767
     }
    },
    {
     "model": "qwen3-asr-arabic-uae",
     "url": "https://huggingface.co/vadimbelsky/qwen3-asr-arabic-uae",
     "kind": "open",
     "v": {
      "clean": 25.49035732646186,
      "noisy": 38.303239108215614,
      "codeswitch": 23.935337404102082,
      "telephony": 37.419840199903895,
      "callcenter": 64.83317358736842,
      "avg": 37.511247428358665
     }
    },
    {
     "model": "whisper-large-v3",
     "url": "https://huggingface.co/openai/whisper-large-v3",
     "kind": "open",
     "v": {
      "clean": 25.58716100162744,
      "noisy": 40.45415438791417,
      "codeswitch": 28.25661499921716,
      "telephony": 40.001225842253795,
      "callcenter": 65.53072682719987,
      "avg": 38.95568154256918
     }
    },
    {
     "model": "qwencleo-asr",
     "url": "https://huggingface.co/mohammedaly22/QwenCleo-ASR",
     "kind": "open",
     "v": {
      "clean": 27.674703177706704,
      "noisy": 44.18383131678806,
      "codeswitch": 28.777203695005483,
      "telephony": 41.42131720939272,
      "callcenter": 77.26961112060975,
      "avg": 42.929626845216596
     }
    },
    {
     "model": "qwen3-asr-0.6b",
     "url": "https://huggingface.co/Qwen/Qwen3-ASR-0.6B-hf",
     "kind": "open",
     "v": {
      "clean": 29.299245581063225,
      "noisy": 40.04507234185063,
      "codeswitch": 28.47189603882887,
      "telephony": 42.89993902464297,
      "callcenter": 67.0527612904048,
      "avg": 41.09922647154217
     }
    },
    {
     "model": "moulsot-v0.3",
     "url": "https://huggingface.co/atlasia/moulsot.v0.3",
     "kind": "open",
     "v": {
      "clean": 37.18439475262826,
      "noisy": 53.82755118239285,
      "codeswitch": 41.71755127602943,
      "telephony": 55.34620849607712,
      "callcenter": 101.55398080750051,
      "avg": 56.640463829097285
     }
    },
    {
     "model": "whisper-large-v3-turbo",
     "url": "https://huggingface.co/openai/whisper-large-v3-turbo",
     "kind": "open",
     "v": {
      "clean": 31.824698531265618,
      "noisy": 41.57062997947442,
      "codeswitch": 24.44026929700955,
      "telephony": 46.792013114109416,
      "callcenter": 90.8066963197441,
      "avg": 47.338007799335784
     }
    },
    {
     "model": "soniox-stt-async-v5",
     "url": "https://soniox.com",
     "kind": "api",
     "v": {
      "clean": 12.535062681952013,
      "noisy": 21.07824822326439,
      "codeswitch": 16.799749491153907,
      "telephony": 19.574023195411286,
      "callcenter": 37.52667571744883,
      "avg": 20.64912390386028
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run3",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run3",
     "kind": "gated",
     "ours": "run3",
     "v": {
      "clean": 10.884656908751909,
      "noisy": 23.59914895002671,
      "codeswitch": 17.187255362454987,
      "telephony": 15.555189669739638,
      "callcenter": 32.45718041276703,
      "avg": 18.670701813314007
     },
     "dag": "Trained on the train split of: cv18_ar, masc_clean, masc_noisy, mgb2, sada"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5",
     "kind": "gated",
     "ours": "run5",
     "v": {
      "clean": 10.782342610026333,
      "noisy": 15.930933990151589,
      "codeswitch": 17.320338186942227,
      "telephony": 15.246331108820007,
      "callcenter": 27.321019804316368,
      "avg": 16.408964013353454
     },
     "dag": "Trained on the train split of: cv18_ar, masc_clean, masc_noisy, mgb2, sada"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5base",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5base",
     "kind": "gated",
     "ours": "run5base",
     "v": {
      "clean": 11.74951097356784,
      "noisy": 16.894604816307716,
      "codeswitch": 19.696258024111476,
      "telephony": 16.07596360133621,
      "callcenter": 34.02018913941972,
      "avg": 18.635375913757237
     },
     "dag": "Trained on the train split of: cv18_ar, masc_clean, masc_noisy, mgb2, sada"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6",
     "url": null,
     "kind": "private",
     "ours": "run6",
     "v": {
      "avg": 16.745246732294017,
      "clean": 10.947374370275858,
      "noisy": 16.616627494599985,
      "codeswitch": 18.45937059652419,
      "telephony": 15.54284675145819,
      "callcenter": 27.333325842817064
     },
     "dag": "Trained on the train split of: cv18_ar, masc_clean, masc_noisy, mgb2, sada"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6-soup50",
     "url": null,
     "kind": "private",
     "ours": "soup-r6-50",
     "v": {
      "avg": 16.41738102422492,
      "clean": 10.677266910759633,
      "noisy": 16.284666874238486,
      "codeswitch": 17.833098481290122,
      "telephony": 15.091905986524838,
      "callcenter": 27.179732644446474
     },
     "dag": "Trained on the train split of: cv18_ar, masc_clean, masc_noisy, mgb2, sada"
    }
   ]
  },
  {
   "id": "lbsets",
   "title": "Leaderboard test sets",
   "sort": "avg",
   "blurb": "The six full test sets of the Open Universal Arabic ASR Leaderboard, run by us (vLLM, repetition repair, the leaderboard's normaliser). This is not a leaderboard submission: our fine-tunes saw the train splits of some of these corpora (†), so they are not eligible there.",
   "links": [
    [
     "Open Universal Arabic ASR Leaderboard",
     "https://huggingface.co/spaces/elmresearchcenter/open_universal_arabic_asr_leaderboard"
    ]
   ],
   "cols": [
    {
     "k": "avg",
     "l": "Average",
     "d": 2,
     "higher": false
    },
    {
     "k": "sada",
     "l": "SADA",
     "d": 2,
     "higher": false
    },
    {
     "k": "common_voice",
     "l": "Common Voice",
     "d": 2,
     "higher": false
    },
    {
     "k": "masc_clean",
     "l": "MASC clean",
     "d": 2,
     "higher": false
    },
    {
     "k": "masc_noisy",
     "l": "MASC noisy",
     "d": 2,
     "higher": false
    },
    {
     "k": "mgb2",
     "l": "MGB-2",
     "d": 2,
     "higher": false
    },
    {
     "k": "casablanca",
     "l": "Casablanca",
     "d": 2,
     "higher": false
    },
    {
     "k": "loops",
     "l": "Loops",
     "d": 0,
     "higher": false
    }
   ],
   "rows": [
    {
     "model": "qwen3-asr-1.7b-ar-run5",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5",
     "kind": "gated",
     "ours": "run5",
     "v": {
      "avg": 17.82,
      "sada": 24.16,
      "common_voice": 7.57,
      "masc_clean": 5.39,
      "masc_noisy": 16.21,
      "mgb2": 9.85,
      "casablanca": 43.72,
      "loops": 2
     },
     "dag": "Trained on the train splits of MASC and Common Voice and on MGB-2 / SADA train rows"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5base",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5base",
     "kind": "gated",
     "ours": "run5base",
     "v": {
      "avg": 18.26,
      "sada": 25.08,
      "common_voice": 8.2,
      "masc_clean": 5.85,
      "masc_noisy": 16.43,
      "mgb2": 9.94,
      "casablanca": 44.08,
      "loops": 6
     },
     "dag": "Trained on the train splits of MASC and Common Voice and on MGB-2 / SADA train rows"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5-comboB",
     "url": null,
     "kind": "private",
     "ours": "comboB",
     "v": {
      "avg": 18.14,
      "sada": 24.53,
      "common_voice": 7.84,
      "masc_clean": 5.51,
      "masc_noisy": 16.41,
      "mgb2": 10.34,
      "casablanca": 44.21,
      "loops": 7
     },
     "dag": "Trained on the train splits of MASC and Common Voice and on MGB-2 / SADA train rows"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run4",
     "url": null,
     "kind": "private",
     "ours": "run4",
     "v": {
      "avg": 17.92,
      "sada": 24.38,
      "common_voice": 7.59,
      "masc_clean": 5.41,
      "masc_noisy": 16.13,
      "mgb2": 10.09,
      "casablanca": 43.91,
      "loops": 9
     },
     "dag": "Trained on the train splits of MASC and Common Voice and on MGB-2 / SADA train rows"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run3",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run3",
     "kind": "gated",
     "ours": "run3",
     "v": {
      "avg": 18.82,
      "sada": 24.7,
      "common_voice": 8.15,
      "masc_clean": 6.01,
      "masc_noisy": 16.78,
      "mgb2": 10.46,
      "casablanca": 46.85,
      "loops": 25
     },
     "dag": "Trained on the train splits of MASC and Common Voice and on MGB-2 / SADA train rows"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run2",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run2",
     "kind": "gated",
     "ours": "run2",
     "v": {
      "avg": 23.04,
      "sada": 24.3,
      "common_voice": 13.49,
      "masc_clean": 17.67,
      "masc_noisy": 25.14,
      "mgb2": 10.36,
      "casablanca": 47.26,
      "loops": 11
     },
     "dag": "Trained on the train splits of MASC and Common Voice and on MGB-2 / SADA train rows"
    },
    {
     "model": "Qwen3-ASR-1.7B",
     "url": "https://huggingface.co/Qwen/Qwen3-ASR-1.7B",
     "kind": "open",
     "ours": "base",
     "v": {
      "avg": 31.88,
      "sada": 42.37,
      "common_voice": 16.25,
      "masc_clean": 21.1,
      "masc_noisy": 33.2,
      "mgb2": 16.58,
      "casablanca": 61.81,
      "loops": 18
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6",
     "url": null,
     "kind": "private",
     "ours": "run6",
     "v": {
      "avg": 18.03,
      "sada": 24.07,
      "common_voice": 7.62,
      "masc_clean": 5.68,
      "masc_noisy": 17.32,
      "mgb2": 10.05,
      "casablanca": 43.45,
      "loops": 2
     },
     "dag": "Trained on the train splits of MASC and Common Voice and on MGB-2 / SADA train rows"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6-soup50",
     "url": null,
     "kind": "private",
     "ours": "soup-r6-50",
     "v": {
      "avg": 17.79,
      "sada": 23.95,
      "common_voice": 7.45,
      "masc_clean": 5.45,
      "masc_noisy": 16.62,
      "mgb2": 9.9,
      "casablanca": 43.35,
      "loops": 1
     },
     "dag": "Trained on the train splits of MASC and Common Voice and on MGB-2 / SADA train rows"
    }
   ]
  },
  {
   "id": "kejue",
   "title": "kejue-slm-eval",
   "sort": "kejue.all_arabic",
   "blurb": "KejueAI's spoken-language evaluation set: ASR, dialect, multi-turn, tool-call, audio-reasoning and translation prompts in Arabic dialects (Sudanese, Yemeni, Maghrebi, Gulf, Levantine, Egyptian) and English. WER (%) on the transcripts; tool_call holds Latin-script entities.",
   "links": [],
   "cols": [
    {
     "k": "kejue.all_arabic",
     "l": "All configs",
     "d": 2,
     "higher": false
    },
    {
     "k": "kejue.asr_arabic",
     "l": "asr",
     "d": 2,
     "higher": false
    },
    {
     "k": "kejue.asr.sudanese",
     "l": "Sudanese",
     "d": 2,
     "higher": false
    },
    {
     "k": "kejue.asr.yemeni",
     "l": "Yemeni",
     "d": 2,
     "higher": false
    },
    {
     "k": "kejue.asr.maghrebi",
     "l": "Maghrebi",
     "d": 2,
     "higher": false
    },
    {
     "k": "kejue.asr.msa",
     "l": "MSA (FLEURS)",
     "d": 2,
     "higher": false
    },
    {
     "k": "kejue.cfg.audio_reasoning",
     "l": "audio_reasoning",
     "d": 2,
     "higher": false
    },
    {
     "k": "kejue.tool_call",
     "l": "tool_call",
     "d": 2,
     "higher": false
    },
    {
     "k": "kejue.english",
     "l": "English",
     "d": 2,
     "higher": false
    }
   ],
   "rows": [
    {
     "model": "qwen3-asr-1.7b-ar-run5",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5",
     "kind": "gated",
     "ours": "run5",
     "v": {
      "kejue.all_arabic": 19.08082540549202,
      "kejue.asr_arabic": 24.502577858716116,
      "kejue.asr.sudanese": 40.201567749160134,
      "kejue.asr.yemeni": 40.5259754257383,
      "kejue.asr.maghrebi": 30.638464168824193,
      "kejue.asr.msa": 4.6692077320991014,
      "kejue.cfg.audio_reasoning": 14.429713889000148,
      "kejue.tool_call": 18.82967607105538,
      "kejue.english": 13.54422746880653
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5base",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5base",
     "kind": "gated",
     "ours": "run5base",
     "v": {
      "kejue.all_arabic": 19.2775624781156,
      "kejue.asr_arabic": 25.078443194822135,
      "kejue.asr.sudanese": 41.881298992161256,
      "kejue.asr.yemeni": 41.88402672989868,
      "kejue.asr.maghrebi": 31.126960089883248,
      "kejue.asr.msa": 4.941464742717125,
      "kejue.cfg.audio_reasoning": 14.293534131720536,
      "kejue.tool_call": 19.456635318704286,
      "kejue.english": 12.998438393864898
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5-comboB",
     "url": null,
     "kind": "private",
     "ours": "comboB",
     "v": {
      "kejue.all_arabic": 18.83156576596894,
      "kejue.asr_arabic": 24.63054793340634,
      "kejue.asr.sudanese": 41.2094064949608,
      "kejue.asr.yemeni": 41.00021556369907,
      "kejue.asr.maghrebi": 30.819207659616044,
      "kejue.asr.msa": 4.53307922679009,
      "kejue.cfg.audio_reasoning": 13.699416563196753,
      "kejue.tool_call": 18.425635667014976,
      "kejue.english": 13.238090820538986
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run4",
     "url": null,
     "kind": "private",
     "ours": "run4",
     "v": {
      "kejue.all_arabic": 19.212873666906034,
      "kejue.asr_arabic": 24.42874896947175,
      "kejue.asr.sudanese": 40.59350503919373,
      "kejue.asr.yemeni": 40.75231730976503,
      "kejue.asr.maghrebi": 30.677543842508914,
      "kejue.asr.msa": 4.451402123604683,
      "kejue.cfg.audio_reasoning": 14.56389100279035,
      "kejue.tool_call": 19.547196098920235,
      "kejue.english": 12.911854291324582
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run3",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run3",
     "kind": "gated",
     "ours": "run3",
     "v": {
      "kejue.all_arabic": 18.85500804159075,
      "kejue.asr_arabic": 25.38237212221142,
      "kejue.asr.sudanese": 44.00895856662934,
      "kejue.asr.yemeni": 41.05410648846734,
      "kejue.asr.maghrebi": 32.78784622148405,
      "kejue.asr.msa": 4.818949087939014,
      "kejue.cfg.audio_reasoning": 13.093950681566335,
      "kejue.tool_call": 20.306513409961685,
      "kejue.english": 15.308378558065463
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6",
     "url": null,
     "kind": "private",
     "ours": "run6",
     "v": {
      "kejue.all_arabic": 19.590323978183847,
      "kejue.asr_arabic": 24.64777467423003,
      "kejue.asr.sudanese": 40.81746920492721,
      "kejue.asr.yemeni": 41.981030394481564,
      "kejue.asr.maghrebi": 30.31605686092521,
      "kejue.asr.msa": 4.546692077320991,
      "kejue.cfg.audio_reasoning": 15.101934553610766,
      "kejue.tool_call": 19.44270289097875,
      "kejue.english": 12.556241013034
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6-soup50",
     "url": null,
     "kind": "private",
     "ours": "soup-r6-50",
     "v": {
      "kejue.all_arabic": 19.251152825832794,
      "kejue.asr_arabic": 24.521035081027208,
      "kejue.asr.sudanese": 40.31354983202687,
      "kejue.asr.yemeni": 40.86009915930157,
      "kejue.asr.maghrebi": 30.570074739875924,
      "kejue.asr.msa": 4.6419820310372994,
      "kejue.cfg.audio_reasoning": 14.746131560326297,
      "kejue.tool_call": 18.718216649251133,
      "kejue.english": 13.106668522040293
     }
    }
   ]
  },
  {
   "id": "codeswitch",
   "title": "Code-switching",
   "sort": "cs.both.wer",
   "blurb": "600 Arabic-English code-switched utterances (Egyptian and Saudi), decoded with Arabic forced. WER (%) after a full normaliser; English-word WER counts the English words alone, and Latin kept is the share of English words written in Latin script (higher is better).",
   "links": [],
   "cols": [
    {
     "k": "cs.both.wer",
     "l": "WER",
     "d": 2,
     "higher": false
    },
    {
     "k": "cs.egyptian.wer",
     "l": "Egyptian",
     "d": 2,
     "higher": false
    },
    {
     "k": "cs.saudi.wer",
     "l": "Saudi",
     "d": 2,
     "higher": false
    },
    {
     "k": "cs.both.en_wer",
     "l": "English-word WER",
     "d": 2,
     "higher": false
    },
    {
     "k": "cs.both.latin",
     "l": "Latin kept %",
     "d": 2,
     "higher": true
    }
   ],
   "rows": [
    {
     "model": "qwen3-asr-1.7b-ar-run5",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5",
     "kind": "gated",
     "ours": "run5",
     "v": {
      "cs.both.wer": 22.010892333472977,
      "cs.egyptian.wer": 20.783719419787833,
      "cs.saudi.wer": 22.785675232367414,
      "cs.both.en_wer": 51.19923249120563,
      "cs.both.latin": 57.14742564758554
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5base",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5base",
     "kind": "gated",
     "ours": "run5base",
     "v": {
      "cs.both.wer": 22.320904901550062,
      "cs.egyptian.wer": 20.04762935700368,
      "cs.saudi.wer": 23.756150902132312,
      "cs.both.en_wer": 51.03933482571154,
      "cs.both.latin": 55.80428525743524
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5-comboB",
     "url": null,
     "kind": "private",
     "ours": "comboB",
     "v": {
      "cs.both.wer": 22.547130289065773,
      "cs.egyptian.wer": 19.28988958649058,
      "cs.saudi.wer": 24.603608529250955,
      "cs.both.en_wer": 53.181963543332266,
      "cs.both.latin": 53.50175887432044
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run4",
     "url": null,
     "kind": "private",
     "ours": "run4",
     "v": {
      "cs.both.wer": 28.051948051948052,
      "cs.egyptian.wer": 30.785884390560728,
      "cs.saudi.wer": 26.325861126298523,
      "cs.both.en_wer": 68.69203709625839,
      "cs.both.latin": 34.98560921010553
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run3",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run3",
     "kind": "gated",
     "ours": "run3",
     "v": {
      "cs.both.wer": 22.23711772098869,
      "cs.egyptian.wer": 13.076423468283178,
      "cs.saudi.wer": 28.02077638053581,
      "cs.both.en_wer": 53.56571794051807,
      "cs.both.latin": 51.29517109050208
     }
    },
    {
     "model": "Qwen3-ASR-1.7B",
     "url": "https://huggingface.co/Qwen/Qwen3-ASR-1.7B",
     "kind": "open",
     "ours": "base",
     "v": {
      "cs.both.wer": 46.63594470046083,
      "cs.egyptian.wer": 51.41805585624594,
      "cs.saudi.wer": 43.61673045379989,
      "cs.both.en_wer": 91.52542372881356,
      "cs.both.latin": 12.951710905020787
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6",
     "url": null,
     "kind": "private",
     "ours": "run6",
     "v": {
      "cs.both.wer": 22.3963133640553,
      "cs.egyptian.wer": 20.480623511582593,
      "cs.saudi.wer": 23.60579551667578,
      "cs.both.en_wer": 52.2865366165654,
      "cs.both.latin": 54.972817396866006
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6-soup50",
     "url": null,
     "kind": "private",
     "ours": "soup-r6-50",
     "v": {
      "cs.both.wer": 22.153330540427316,
      "cs.egyptian.wer": 20.69712058887205,
      "cs.saudi.wer": 23.072717331875342,
      "cs.both.en_wer": 52.2865366165654,
      "cs.both.latin": 55.38855132715062
     }
    }
   ]
  },
  {
   "id": "testarabic",
   "title": "test-arabic-dataset",
   "sort": "wer",
   "blurb": "552 speech segments (78 min, 11,819 reference words) of hard call-centre, dialect and English recordings from kadirnar/test-arabic-dataset; the references are reviewed transcripts in the client's written format. Arabic and English columns split the segments by script; Reliable keeps the segments the reviewers marked certain.",
   "links": [
    [
     "Dataset",
     "https://huggingface.co/datasets/kadirnar/test-arabic-dataset"
    ],
    [
     "Benchmark Space",
     "https://huggingface.co/spaces/kadirnar/test-arabic-asr-benchmark"
    ]
   ],
   "cols": [
    {
     "k": "wer",
     "l": "WER",
     "d": 2,
     "higher": false
    },
    {
     "k": "ar",
     "l": "Arabic",
     "d": 2,
     "higher": false
    },
    {
     "k": "en",
     "l": "English",
     "d": 2,
     "higher": false
    },
    {
     "k": "rel",
     "l": "Reliable",
     "d": 2,
     "higher": false
    },
    {
     "k": "cer",
     "l": "CER",
     "d": 2,
     "higher": false
    },
    {
     "k": "noise",
     "l": "Words on silence",
     "d": 0,
     "higher": false
    }
   ],
   "rows": [
    {
     "model": "soniox-stt-async-v5",
     "url": "https://soniox.com",
     "kind": "api",
     "v": {
      "wer": 11.016160419663255,
      "ar": 17.909496494582537,
      "en": 5.456852791878172,
      "rel": 9.104422147900408,
      "cer": 6.862828327419053,
      "noise": 81
     },
     "ci": [
      9.763369497609423,
      12.392121331799071
     ]
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6b-soup70",
     "url": null,
     "kind": "private",
     "v": {
      "wer": 11.058465183179626,
      "ar": 15.338857021457402,
      "en": 6.782289904117316,
      "rel": 8.67707172054998,
      "cer": 7.258434215664635,
      "noise": 119
     },
     "ci": [
      9.873091874597083,
      12.354932798130081
     ]
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6-soup50",
     "url": null,
     "kind": "private",
     "v": {
      "wer": 11.14307471021237,
      "ar": 15.785001062247716,
      "en": 6.725888324873096,
      "rel": 8.862876254180602,
      "cer": 7.307672716776746,
      "noise": 113
     },
     "ci": [
      9.945732765903829,
      12.480839933607053
     ],
     "ours": "soup-r6-50"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6",
     "url": null,
     "kind": "private",
     "v": {
      "wer": 11.68457568322193,
      "ar": 16.146165285744633,
      "en": 7.402707275803723,
      "rel": 9.057971014492754,
      "cer": 7.633665551725895,
      "noise": 112
     },
     "ci": [
      10.369098935793392,
      13.034691075751084
     ],
     "ours": "run6"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5-comboB",
     "url": null,
     "kind": "private",
     "v": {
      "wer": 11.777646162957948,
      "ar": 15.338857021457402,
      "en": 8.192329385222786,
      "rel": 9.819769602378297,
      "cer": 7.908722006214239,
      "noise": 103
     },
     "ci": [
      10.604955993659692,
      12.982849501859931
     ],
     "ours": "comboB"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5base",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5base",
     "kind": "gated",
     "v": {
      "wer": 11.786107115661224,
      "ar": 16.33736987465477,
      "en": 7.4450084602368864,
      "rel": 9.596804162021554,
      "cer": 7.9002326094707715,
      "noise": 108
     },
     "ci": [
      10.509582384979597,
      13.279968588326934
     ],
     "ours": "run5base"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5",
     "kind": "gated",
     "v": {
      "wer": 11.963787122429984,
      "ar": 15.785001062247716,
      "en": 8.305132543711224,
      "rel": 9.940542549238202,
      "cer": 8.04625023345841,
      "noise": 97
     },
     "ci": [
      10.659891070584782,
      13.288108566175408
     ],
     "ours": "run5"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run4",
     "url": null,
     "kind": "private",
     "v": {
      "wer": 11.980709027836536,
      "ar": 17.01720841300191,
      "en": 7.473209249858996,
      "rel": 9.875510962467485,
      "cer": 7.827223797476951,
      "noise": 100
     },
     "ci": [
      10.639701412912062,
      13.335118869467003
     ],
     "ours": "run4"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run2",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run2",
     "kind": "gated",
     "v": {
      "wer": 16.355021575429394,
      "ar": 18.716804758869767,
      "en": 13.324873096446701,
      "rel": 14.037532515793385,
      "cer": 11.861385129972664,
      "noise": 109
     },
     "ci": [
      14.774511789686986,
      18.05832309625676
     ],
     "ours": "run2"
    },
    {
     "model": "qwen3-asr-1.7b-ar-run3",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run3",
     "kind": "gated",
     "v": {
      "wer": 16.49039681868178,
      "ar": 18.971744210749947,
      "en": 13.381274675690918,
      "rel": 14.195466369379414,
      "cer": 11.929300303920403,
      "noise": 112
     },
     "ci": [
      14.775755829730636,
      18.478541329946893
     ],
     "ours": "run3"
    },
    {
     "model": "Qwen3-ASR-1.7B",
     "url": "https://huggingface.co/Qwen/Qwen3-ASR-1.7B",
     "kind": "open",
     "v": {
      "wer": 19.079448345883744,
      "ar": 33.92819205438708,
      "en": 8.220530174844894,
      "rel": 17.08472686733556,
      "cer": 11.783282679932764,
      "noise": 73
     },
     "ci": [
      16.870097855599717,
      21.607566380872317
     ],
     "ours": "base"
    }
   ]
  },
  {
   "id": "smartturn",
   "title": "smart-turn Arabic",
   "sort": "wer",
   "blurb": "7,396 Arabic clips (17.6 h) of smart-turn v3.1, synthetic speech. References: ROVER consensus of 7 pre-registered open models; a voter is scored against the consensus of the other six, every other model against all seven leave-one-out consensuses. WER after the extended normaliser; the leaderboard-normaliser WER is shown too. ⚠ marks a fact that matters for reading the score.",
   "links": [
    [
     "smart-turn benchmark Space",
     "https://huggingface.co/spaces/kadirnar/smart-turn-arabic-asr-benchmark"
    ],
    [
     "Predictions and code",
     "https://huggingface.co/datasets/kadirnar/smart-turn-arabic-asr-benchmark"
    ]
   ],
   "cols": [
    {
     "k": "wer",
     "l": "WER",
     "d": 2,
     "higher": false
    },
    {
     "k": "cer",
     "l": "CER",
     "d": 2,
     "higher": false
    },
    {
     "k": "elm",
     "l": "WER (leaderboard norm.)",
     "d": 2,
     "higher": false
    },
    {
     "k": "ins",
     "l": "Insertions",
     "d": 2,
     "higher": false
    },
    {
     "k": "runaway",
     "l": "Runaway %",
     "d": 2,
     "higher": false
    }
   ],
   "rows": [
    {
     "model": "mistralai/Voxtral-Mini-4B-Realtime-Arabic",
     "url": "https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-Arabic",
     "kind": "open",
     "v": {
      "wer": 4.1565384605,
      "cer": 1.50055554,
      "elm": 5.9818198456,
      "ins": 0.6178742753,
      "runaway": 0.0173839141
     },
     "ci": [
      3.9604780404,
      4.3635298703
     ],
     "voter": false
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6-soup50",
     "url": null,
     "kind": "private",
     "ours": "soup-r6-50",
     "v": {
      "wer": 4.6140799035,
      "cer": 1.8478542182,
      "elm": 7.5981342981,
      "ins": 0.7802692664,
      "runaway": 0.005794638
     },
     "ci": [
      4.4091394612,
      4.8408020692
     ],
     "voter": false,
     "warn": "Weight average of run 6 and run 5 (50/50). Training data includes synthetic speech from the same TTS voice family as this test set (Google Chirp 3 HD voices, e.g. Zephyr/Puck): an acoustic-domain advantage; the 8-gram text check covered the synthetic sets of runs 2-3 only. Run with vLLM (greedy, forced Arabic, the same 256-token cap, no repetition repair) on an H100, not with this benchmark's transformers runner: on run 3 the two runners give the same text on 94.9 % of clips and 5.43 vs 5.37 WER."
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6",
     "url": null,
     "kind": "private",
     "ours": "run6",
     "v": {
      "wer": 4.6890909232,
      "cer": 1.9125791076,
      "elm": 6.8298661814,
      "ins": 0.8296321882,
      "runaway": 0.0366993742
     },
     "ci": [
      4.4509594699,
      4.9767694211
     ],
     "voter": false,
     "warn": "Training data includes synthetic speech from the same TTS voice family as this test set (Google Chirp 3 HD voices, e.g. Zephyr/Puck): an acoustic-domain advantage; the 8-gram text check covered the synthetic sets of runs 2-3 only. Run with vLLM (greedy, forced Arabic, the same 256-token cap, no repetition repair) on an H100, not with this benchmark's transformers runner: on run 3 the two runners give the same text on 94.9 % of clips and 5.43 vs 5.37 WER."
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5base",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5base",
     "kind": "gated",
     "ours": "run5base",
     "v": {
      "wer": 4.7134501718,
      "cer": 1.9265308149,
      "elm": 6.6008421177,
      "ins": 0.8313076921,
      "runaway": 0.0289731901
     },
     "ci": [
      4.4271652544,
      5.0584806229
     ],
     "voter": false,
     "warn": "Training data includes synthetic speech from the same TTS voice family as this test set (Google Chirp 3 HD voices, e.g. Zephyr/Puck): an acoustic-domain advantage; the 8-gram text check covered the synthetic sets of runs 2-3 only. Run with vLLM (greedy, forced Arabic, the same 256-token cap, no repetition repair) on an H100, not with this benchmark's transformers runner: on run 3 the two runners give the same text on 94.9 % of clips and 5.43 vs 5.37 WER."
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5",
     "kind": "gated",
     "ours": "run5",
     "v": {
      "wer": 4.8642455206,
      "cer": 1.9802000679,
      "elm": 8.8303791334,
      "ins": 0.9926716038,
      "runaway": 0.0444255582
     },
     "ci": [
      4.5347610902,
      5.2535447889
     ],
     "voter": false,
     "warn": "Training data includes synthetic speech from the same TTS voice family as this test set (Google Chirp 3 HD voices, e.g. Zephyr/Puck): an acoustic-domain advantage; the 8-gram text check covered the synthetic sets of runs 2-3 only. Run with vLLM (greedy, forced Arabic, the same 256-token cap, no repetition repair) on an H100, not with this benchmark's transformers runner: on run 3 the two runners give the same text on 94.9 % of clips and 5.43 vs 5.37 WER."
    },
    {
     "model": "Qwen/Qwen3-Omni-30B-A3B-Instruct",
     "url": "https://huggingface.co/Qwen/Qwen3-Omni-30B-A3B-Instruct",
     "kind": "open",
     "v": {
      "wer": 4.8990444473,
      "cer": 2.0269739138,
      "elm": 13.4011942245,
      "ins": 0.6750991769,
      "runaway": 0.005794638
     },
     "ci": [
      4.7085458635,
      5.1150568335
     ],
     "voter": false
    },
    {
     "model": "google/gemma-4-E4B-it",
     "url": "https://huggingface.co/google/gemma-4-E4B-it",
     "kind": "open",
     "v": {
      "wer": 4.9797812037,
      "cer": 1.8996479808,
      "elm": 18.0559564182,
      "ins": 1.0317001841,
      "runaway": 0.0405624662
     },
     "ci": [
      4.7243274055,
      5.2624305308
     ],
     "voter": true
    },
    {
     "model": "CohereLabs/cohere-transcribe-arabic-07-2026",
     "url": "https://huggingface.co/CohereLabs/cohere-transcribe-arabic-07-2026",
     "kind": "open",
     "v": {
      "wer": 5.0191864927,
      "cer": 1.9655564838,
      "elm": 9.5051883427,
      "ins": 1.1114622365,
      "runaway": 0.0405624662
     },
     "ci": [
      4.7744943072,
      5.2802711166
     ],
     "voter": true
    },
    {
     "model": "qwen3-asr-1.7b-ar-run2",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run2",
     "kind": "gated",
     "ours": "run2",
     "v": {
      "wer": 5.1890355026,
      "cer": 1.9982430502,
      "elm": 15.286357409,
      "ins": 0.9141806915,
      "runaway": 0.0328362822
     },
     "ci": [
      4.871699933,
      5.5711081336
     ],
     "voter": false,
     "warn": "Training data includes synthetic speech from the same TTS voice family as this test set (Google Chirp 3 HD voices, e.g. Zephyr/Puck): an acoustic-domain advantage. No text overlap found: 0 shared 8-grams with the benchmark across 551k synthetic training utterances."
    },
    {
     "model": "qwen3-asr-1.7b-ar-run3",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run3",
     "kind": "gated",
     "ours": "run3",
     "v": {
      "wer": 5.3679277626,
      "cer": 2.1511142289,
      "elm": 21.9664622162,
      "ins": 0.8769329515,
      "runaway": 0.0482886502
     },
     "ci": [
      5.0611689314,
      5.7253743105
     ],
     "voter": false,
     "warn": "Training data includes synthetic speech from the same TTS voice family as this test set (Google Chirp 3 HD voices, e.g. Zephyr/Puck): an acoustic-domain advantage. No text overlap found: 0 shared 8-grams with the benchmark across 551k synthetic training utterances."
    },
    {
     "model": "facebook/omniASR-LLM-7B",
     "url": "https://huggingface.co/facebook/omniASR-LLM-7B",
     "kind": "open",
     "v": {
      "wer": 5.4109138132,
      "cer": 2.479661914,
      "elm": 6.892513894,
      "ins": 0.7963061495,
      "runaway": 0.0270416441
     },
     "ci": [
      5.1638867813,
      5.6670354854
     ],
     "voter": true
    },
    {
     "model": "openai/whisper-large-v3",
     "url": "https://huggingface.co/openai/whisper-large-v3",
     "kind": "open",
     "v": {
      "wer": 5.6510480109,
      "cer": 2.2656221088,
      "elm": 7.5714221239,
      "ins": 1.0268070631,
      "runaway": 0.0270416441
     },
     "ci": [
      5.3438958313,
      5.9896424545
     ],
     "voter": true
    },
    {
     "model": "audarai/Audar-ASR-V1-Turbo",
     "url": "https://huggingface.co/audarai/Audar-ASR-V1-Turbo",
     "kind": "open",
     "v": {
      "wer": 5.6511942316,
      "cer": 2.3017104808,
      "elm": 9.1767504302,
      "ins": 0.995944119,
      "runaway": 0.0405624662
     },
     "ci": [
      5.3351976178,
      6.048996931
     ],
     "voter": true
    },
    {
     "model": "mistralai/Voxtral-Mini-4B-Realtime-2602",
     "url": "https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602",
     "kind": "open",
     "v": {
      "wer": 6.0762792472,
      "cer": 2.3007675508,
      "elm": 7.7065244148,
      "ins": 0.8603067976,
      "runaway": 0.0173839141
     },
     "ci": [
      5.8006999194,
      6.379349913
     ],
     "voter": false
    },
    {
     "model": "openai/whisper-large-v3-turbo",
     "url": "https://huggingface.co/openai/whisper-large-v3-turbo",
     "kind": "open",
     "v": {
      "wer": 6.1622454845,
      "cer": 2.5099512219,
      "elm": 7.7288210738,
      "ins": 1.5834800473,
      "runaway": 0.0772618404
     },
     "ci": [
      5.6951227865,
      6.8095391618
     ],
     "voter": false
    },
    {
     "model": "mistralai/Voxtral-Small-24B-2507",
     "url": "https://huggingface.co/mistralai/Voxtral-Small-24B-2507",
     "kind": "open",
     "v": {
      "wer": 6.3970303363,
      "cer": 2.6657752893,
      "elm": 7.8625472888,
      "ins": 0.7433169053,
      "runaway": 0.0135208221
     },
     "ci": [
      6.0484406422,
      6.8013595052
     ],
     "voter": true
    },
    {
     "model": "Qwen3-ASR-1.7B",
     "url": "https://huggingface.co/Qwen/Qwen3-ASR-1.7B",
     "kind": "open",
     "ours": "base",
     "v": {
      "wer": 6.5403938207,
      "cer": 2.2525686541,
      "elm": 14.3433247283,
      "ins": 0.9125051876,
      "runaway": 0.0405624662
     },
     "ci": [
      6.3008028651,
      6.7861846706
     ],
     "voter": false
    },
    {
     "model": "nvidia/stt_ar_fastconformer_hybrid_large_pcd_v1.0",
     "url": "https://huggingface.co/nvidia/stt_ar_fastconformer_hybrid_large_pcd_v1.0",
     "kind": "open",
     "v": {
      "wer": 6.5860190801,
      "cer": 2.6432624286,
      "elm": 18.2988551374,
      "ins": 1.1281296479,
      "runaway": 0.0347678282
     },
     "ci": [
      6.3330067161,
      6.8524940018
     ],
     "voter": false
    },
    {
     "model": "nvidia/stt_ar_fastconformer_hybrid_large_pc_v1.0",
     "url": "https://huggingface.co/nvidia/stt_ar_fastconformer_hybrid_large_pc_v1.0",
     "kind": "open",
     "v": {
      "wer": 6.7185127712,
      "cer": 2.6516748255,
      "elm": 18.4437189796,
      "ins": 1.217446893,
      "runaway": 0.0270416441
     },
     "ci": [
      6.4579165999,
      6.9651134298
     ],
     "voter": false
    },
    {
     "model": "nvidia/nemotron-3.5-asr-streaming-0.6b",
     "url": "https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b",
     "kind": "open",
     "v": {
      "wer": 6.9293871061,
      "cer": 2.4367898265,
      "elm": 17.7031197676,
      "ins": 1.4283391154,
      "runaway": 0.0676041103
     },
     "ci": [
      6.6789092851,
      7.2073718473
     ],
     "voter": true
    },
    {
     "model": "facebook/seamless-m4t-v2-large",
     "url": "https://huggingface.co/facebook/seamless-m4t-v2-large",
     "kind": "open",
     "v": {
      "wer": 7.2531273924,
      "cer": 2.9415117767,
      "elm": 9.1610914279,
      "ins": 0.9897072508,
      "runaway": 0.0540832883
     },
     "ci": [
      6.7640666616,
      7.8966993416
     ],
     "voter": false
    },
    {
     "model": "openai/whisper-large-v2",
     "url": "https://huggingface.co/openai/whisper-large-v2",
     "kind": "open",
     "v": {
      "wer": 7.5896459016,
      "cer": 3.2680093097,
      "elm": 8.4527536374,
      "ins": 1.7024408225,
      "runaway": 0.0714672023
     },
     "ci": [
      6.7893447251,
      8.673531838
     ],
     "voter": false
    },
    {
     "model": "facebook/omniASR-LLM-7B v2",
     "url": "https://github.com/facebookresearch/omnilingual-asr",
     "kind": "open",
     "v": {
      "wer": 7.7481743452,
      "cer": 4.0781966763,
      "elm": 8.4728592952,
      "ins": 1.0885619795,
      "runaway": 0.0309047362
     },
     "ci": [
      7.4502152091,
      8.0466499881
     ],
     "voter": false,
     "warn": "omniASR v2 checkpoints often omit numbers entirely (e.g. 'عام 2008' -> 'عام'): digits appear in ~1.2k clips vs 2.7k for v1. Verified to be model output (official pipeline, tokenizer round-trips digits)."
    },
    {
     "model": "audarai/Audar-ASR-V1-Flash",
     "url": "https://huggingface.co/audarai/Audar-ASR-V1-Flash",
     "kind": "open",
     "v": {
      "wer": 8.0440941066,
      "cer": 2.917400836,
      "elm": 9.1577404849,
      "ins": 0.7044849372,
      "runaway": 0.0193154601
     },
     "ci": [
      7.8071496394,
      8.2974492995
     ],
     "voter": false
    },
    {
     "model": "facebook/omniASR-LLM-1B v2",
     "url": "https://github.com/facebookresearch/omnilingual-asr",
     "kind": "open",
     "v": {
      "wer": 8.3861546671,
      "cer": 4.1950968138,
      "elm": 9.3937241994,
      "ins": 1.6293630765,
      "runaway": 0.0791933864
     },
     "ci": [
      8.0431023309,
      8.6997345119
     ],
     "voter": false,
     "warn": "omniASR v2 checkpoints often omit numbers entirely (e.g. 'عام 2008' -> 'عام'): digits appear in ~1.2k clips vs 2.7k for v1. Verified to be model output (official pipeline, tokenizer round-trips digits)."
    },
    {
     "model": "facebook/omniASR-LLM-3B v2",
     "url": "https://github.com/facebookresearch/omnilingual-asr",
     "kind": "open",
     "v": {
      "wer": 8.5049865573,
      "cer": 4.3245925619,
      "elm": 9.532917215,
      "ins": 1.588506559,
      "runaway": 0.0927142085
     },
     "ci": [
      8.1272830566,
      8.9058983534
     ],
     "voter": false,
     "warn": "omniASR v2 checkpoints often omit numbers entirely (e.g. 'عام 2008' -> 'عام'): digits appear in ~1.2k clips vs 2.7k for v1. Verified to be model output (official pipeline, tokenizer round-trips digits)."
    },
    {
     "model": "google/gemma-4-E2B-it",
     "url": "https://huggingface.co/google/gemma-4-E2B-it",
     "kind": "open",
     "v": {
      "wer": 8.9666523175,
      "cer": 5.125080389,
      "elm": 21.1316907698,
      "ins": 1.8103174951,
      "runaway": 0.0579463803
     },
     "ci": [
      8.2408549686,
      9.7108315344
     ],
     "voter": false
    },
    {
     "model": "oddadmix/whisper-large-v3-turbo-arabic-dialectal",
     "url": "https://huggingface.co/oddadmix/whisper-large-v3-turbo-arabic-dialectal",
     "kind": "open",
     "v": {
      "wer": 9.6470357759,
      "cer": 3.0194298785,
      "elm": 13.1537399746,
      "ins": 2.1777683835,
      "runaway": 0.0328362822
     },
     "ci": [
      9.3555288701,
      9.9627464283
     ],
     "voter": false
    },
    {
     "model": "openai/whisper-medium",
     "url": "https://huggingface.co/openai/whisper-medium",
     "kind": "open",
     "v": {
      "wer": 9.6485823948,
      "cer": 3.7701788347,
      "elm": 11.6070220299,
      "ins": 1.2392284434,
      "runaway": 0.0251100981
     },
     "ci": [
      9.3477202231,
      9.962426846
     ],
     "voter": false
    },
    {
     "model": "MohamedRashad/Arabic-Whisper-CodeSwitching-Edition",
     "url": "https://huggingface.co/MohamedRashad/Arabic-Whisper-CodeSwitching-Edition",
     "kind": "open",
     "v": {
      "wer": 10.78083636,
      "cer": 5.3082453901,
      "elm": 14.1881502924,
      "ins": 2.0035159804,
      "runaway": 0.0521517423
     },
     "ci": [
      10.3546593925,
      11.2814099568
     ],
     "voter": false
    },
    {
     "model": "mistralai/Voxtral-Mini-3B-2507",
     "url": "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507",
     "kind": "open",
     "v": {
      "wer": 11.9865547258,
      "cer": 4.6253472412,
      "elm": 12.7482758754,
      "ins": 1.8616136907,
      "runaway": 0.1313451286
     },
     "ci": [
      11.2095054286,
      12.9267513732
     ],
     "voter": false
    },
    {
     "model": "facebook/omniASR-CTC-7B v2",
     "url": "https://github.com/facebookresearch/omnilingual-asr",
     "kind": "open",
     "v": {
      "wer": 12.0433929727,
      "cer": 5.2565758051,
      "elm": 11.9811677005,
      "ins": 0.6232874417,
      "runaway": 0.0
     },
     "ci": [
      11.6917679698,
      12.3901907763
     ],
     "voter": false,
     "warn": "omniASR v2 checkpoints often omit numbers entirely (e.g. 'عام 2008' -> 'عام'): digits appear in ~1.2k clips vs 2.7k for v1. Verified to be model output (official pipeline, tokenizer round-trips digits)."
    },
    {
     "model": "CohereLabs/cohere-transcribe-03-2026",
     "url": "https://huggingface.co/CohereLabs/cohere-transcribe-03-2026",
     "kind": "open",
     "v": {
      "wer": 12.6858842665,
      "cer": 8.6745602224,
      "elm": 32.2452220064,
      "ins": 1.3088262966,
      "runaway": 0.0193154601
     },
     "ci": [
      12.1615046151,
      13.2517775006
     ],
     "voter": false
    },
    {
     "model": "Qwen/Qwen3-ASR-0.6B-hf",
     "url": "https://huggingface.co/Qwen/Qwen3-ASR-0.6B-hf",
     "kind": "open",
     "v": {
      "wer": 13.1600518633,
      "cer": 4.3131002165,
      "elm": 22.9029218934,
      "ins": 1.5152999281,
      "runaway": 0.0405624662
     },
     "ci": [
      12.8318947959,
      13.5021582955
     ],
     "voter": false
    },
    {
     "model": "facebook/omniASR-CTC-1B v2",
     "url": "https://github.com/facebookresearch/omnilingual-asr",
     "kind": "open",
     "v": {
      "wer": 14.0674016544,
      "cer": 5.6395697082,
      "elm": 14.1963987674,
      "ins": 0.7401860583,
      "runaway": 0.0
     },
     "ci": [
      13.7153291523,
      14.4153033661
     ],
     "voter": false,
     "warn": "omniASR v2 checkpoints often omit numbers entirely (e.g. 'عام 2008' -> 'عام'): digits appear in ~1.2k clips vs 2.7k for v1. Verified to be model output (official pipeline, tokenizer round-trips digits)."
    },
    {
     "model": "asafaya/hubert-large-arabic-transcribe",
     "url": "https://huggingface.co/asafaya/hubert-large-arabic-transcribe",
     "kind": "open",
     "v": {
      "wer": 14.4679759655,
      "cer": 4.5733558706,
      "elm": 24.6425768236,
      "ins": 2.406539105,
      "runaway": 0.0251100981
     },
     "ci": [
      14.1430311278,
      14.8098139935
     ],
     "voter": false
    },
    {
     "model": "UsefulSensors/moonshine-tiny-ar",
     "url": "https://huggingface.co/UsefulSensors/moonshine-tiny-ar",
     "kind": "open",
     "v": {
      "wer": 15.1198758581,
      "cer": 7.8118528373,
      "elm": 17.2900924216,
      "ins": 3.2239272264,
      "runaway": 0.202812331
     },
     "ci": [
      14.5201881368,
      15.8054932881
     ],
     "voter": false
    },
    {
     "model": "openai/whisper-small",
     "url": "https://huggingface.co/openai/whisper-small",
     "kind": "open",
     "v": {
      "wer": 17.6841700972,
      "cer": 6.2642706199,
      "elm": 19.977806447,
      "ins": 2.5109358849,
      "runaway": 0.0733987484
     },
     "ci": [
      17.2300276937,
      18.2031492239
     ],
     "voter": false
    },
    {
     "model": "UsefulSensors/moonshine-base-ar",
     "url": "https://huggingface.co/UsefulSensors/moonshine-base-ar",
     "kind": "open",
     "v": {
      "wer": 18.325114772,
      "cer": 10.6627589628,
      "elm": 21.3594260092,
      "ins": 3.9452960873,
      "runaway": 0.1390713127
     },
     "ci": [
      17.7555491705,
      18.9017542433
     ],
     "voter": false
    },
    {
     "model": "facebook/mms-1b-all",
     "url": "https://huggingface.co/facebook/mms-1b-all",
     "kind": "open",
     "v": {
      "wer": 23.0843190881,
      "cer": 7.2805157397,
      "elm": 24.6293019341,
      "ins": 4.0170849841,
      "runaway": 0.0251100981
     },
     "ci": [
      22.6547922975,
      23.4849786581
     ],
     "voter": false
    },
    {
     "model": "whitefox123/w2v-bert-2.0-arabic-4",
     "url": "https://huggingface.co/whitefox123/w2v-bert-2.0-arabic-4",
     "kind": "open",
     "v": {
      "wer": 27.2715321581,
      "cer": 8.7662001847,
      "elm": 33.695664782,
      "ins": 3.8590720802,
      "runaway": 0.0405624662
     },
     "ci": [
      26.7772341231,
      27.7424017055
     ],
     "voter": false
    },
    {
     "model": "jonatasgrosman/wav2vec2-large-xlsr-53-arabic",
     "url": "https://huggingface.co/jonatasgrosman/wav2vec2-large-xlsr-53-arabic",
     "kind": "open",
     "v": {
      "wer": 32.2421077323,
      "cer": 11.7747583045,
      "elm": 35.7153812149,
      "ins": 6.7152906484,
      "runaway": 0.0173839141
     },
     "ci": [
      31.707842739,
      32.7601515614
     ],
     "voter": false
    },
    {
     "model": "speechbrain/asr-wav2vec2-commonvoice-14-ar",
     "url": "https://huggingface.co/speechbrain/asr-wav2vec2-commonvoice-14-ar",
     "kind": "open",
     "v": {
      "wer": 43.7149271929,
      "cer": 16.8837494099,
      "elm": 46.3400614254,
      "ins": 8.3629553826,
      "runaway": 0.0270416441
     },
     "ci": [
      43.1243648158,
      44.2635112898
     ],
     "voter": false
    },
    {
     "model": "microsoft/VibeVoice-ASR-HF",
     "url": "https://huggingface.co/microsoft/VibeVoice-ASR-HF",
     "kind": "open",
     "v": {
      "wer": 97.1600209309,
      "cer": 66.9245242514,
      "elm": 95.0904819043,
      "ins": 48.7059954684,
      "runaway": 3.8244610987
     },
     "ci": [
      91.8271258404,
      102.7308747108
     ],
     "voter": false
    }
   ]
  },
  {
   "id": "english",
   "title": "English Arena",
   "sort": "avg",
   "blurb": "The Arabic Arena's English suite, run with the arena's runner: clean, accented, added-noise, telephony and call-center conditions. The average is over all conditions (subset WER, %).",
   "links": [
    [
     "Arabic Arena (GitHub)",
     "https://github.com/kadirnar/arabic-arena"
    ]
   ],
   "cols": [
    {
     "k": "avg",
     "l": "Average",
     "d": 2,
     "higher": false
    },
    {
     "k": "clean",
     "l": "Clean",
     "d": 2,
     "higher": false
    },
    {
     "k": "accented",
     "l": "Accented",
     "d": 2,
     "higher": false
    },
    {
     "k": "noisy",
     "l": "Noisy",
     "d": 2,
     "higher": false
    },
    {
     "k": "telephony",
     "l": "Telephony",
     "d": 2,
     "higher": false
    },
    {
     "k": "callcenter",
     "l": "Call-center",
     "d": 2,
     "higher": false
    }
   ],
   "rows": [
    {
     "model": "Qwen3-ASR-1.7B",
     "url": "https://huggingface.co/Qwen/Qwen3-ASR-1.7B",
     "kind": "open",
     "ours": "base",
     "v": {
      "avg": 9.641583047808838,
      "clean": 6.057538195977934,
      "accented": 14.150167279109308,
      "noisy": 8.669470685093769,
      "telephony": 6.541329685258136,
      "callcenter": 17.963184592506448
     }
    },
    {
     "model": "parakeet-tdt-0.6b-v3",
     "url": "https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3",
     "kind": "open",
     "v": {
      "avg": 11.212329334498584,
      "clean": 6.770173319391818,
      "accented": 15.153841934953109,
      "noisy": 9.882851119247132,
      "telephony": 7.8356938642646385,
      "callcenter": 21.509412393854127
     }
    },
    {
     "model": "canary-1b-v2",
     "url": "https://huggingface.co/nvidia/canary-1b-v2",
     "kind": "open",
     "v": {
      "avg": 11.79339085486568,
      "clean": 7.460780003213509,
      "accented": 18.954642680853397,
      "noisy": 10.431327493910446,
      "telephony": 7.850380887678597,
      "callcenter": 21.792670648373466
     }
    },
    {
     "model": "whisper-large-v3",
     "url": "https://huggingface.co/openai/whisper-large-v3",
     "kind": "open",
     "v": {
      "avg": 12.446130752731671,
      "clean": 8.07185126363647,
      "accented": 19.179509680250096,
      "noisy": 10.312800014162757,
      "telephony": 8.262860670671046,
      "callcenter": 23.3597496596489
     }
    },
    {
     "model": "cohere-transcribe-03-2026",
     "url": "https://huggingface.co/CohereLabs/cohere-transcribe-03-2026",
     "kind": "open",
     "v": {
      "avg": 11.408865020069054,
      "clean": 5.899026999016037,
      "accented": 19.77732682498766,
      "noisy": 10.621864519855558,
      "telephony": 6.8028894140845955,
      "callcenter": 23.250708591185294
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5",
     "kind": "gated",
     "ours": "run5",
     "v": {
      "avg": 10.787542312798355,
      "clean": 7.237548992532259,
      "accented": 17.824823122909013,
      "noisy": 9.530170953833505,
      "telephony": 8.86769267982414,
      "callcenter": 17.504011169916247
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5base",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run5base",
     "kind": "gated",
     "ours": "run5base",
     "v": {
      "avg": 10.240182650032354,
      "clean": 7.026604344132302,
      "accented": 16.963747052048483,
      "noisy": 9.004865627944342,
      "telephony": 8.20290673447526,
      "callcenter": 16.605777411266843
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run5-comboB",
     "url": null,
     "kind": "private",
     "ours": "comboB",
     "v": {
      "avg": 10.793840727720923,
      "clean": 7.1888118054263614,
      "accented": 17.139253002797126,
      "noisy": 9.826622589240072,
      "telephony": 8.554829196116247,
      "callcenter": 17.87539550297633
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run4",
     "url": null,
     "kind": "private",
     "ours": "run4",
     "v": {
      "avg": 11.12928336079596,
      "clean": 7.332790217583859,
      "accented": 18.14841221960182,
      "noisy": 10.072183520597768,
      "telephony": 8.947211755793287,
      "callcenter": 18.361937780081607
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run3",
     "url": "https://huggingface.co/Arabic-STT/qwen3-asr-1.7b-ar-run3",
     "kind": "gated",
     "ours": "run3",
     "v": {
      "avg": 19.719087125956708,
      "clean": 10.900604187521077,
      "accented": 19.51955245982559,
      "noisy": 19.780215862006546,
      "telephony": 13.31425973847403,
      "callcenter": 39.84514318611294
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6",
     "url": null,
     "kind": "private",
     "ours": "run6",
     "v": {
      "avg": 9.324433279704724,
      "clean": 6.358935529950931,
      "accented": 15.411616300115174,
      "noisy": 7.539152532376171,
      "telephony": 7.163871135371259,
      "callcenter": 15.935471750671917
     }
    },
    {
     "model": "qwen3-asr-1.7b-ar-run6-soup50",
     "url": null,
     "kind": "private",
     "ours": "soup-r6-50",
     "v": {
      "avg": 9.519266746178067,
      "clean": 6.418543095910691,
      "accented": 15.66390610431635,
      "noisy": 7.800092093599599,
      "telephony": 7.4017404713272175,
      "callcenter": 16.240781943577527
     }
    }
   ]
  }
 ]
}