Upload from GitHub Actions: results: update committed snapshots to current 71-model state

The repo snapshots were stale (40-model b7b017f restore) while HF/the live
dashboard are at 71 fully-covered models. Refresh results.json/models.json/
languages.json/model_health.json so GitHub matches what's published. (Actions
runs push to HF but don't commit back, so these are updated by hand.)

results/languages.json

@@ -7,7 +7,7 @@
     "family":"Indo-European",
     "flores_path":"eng_Latn",
     "fleurs_tag":"en_us",
-    "commonvoice_hours":2714.0,
+    "commonvoice_hours":2744.0,
     "commonvoice_locale":"en",
     "in_benchmark":true
   },
@@ -19,7 +19,7 @@
     "family":"Sino-Tibetan",
     "flores_path":"cmn_Hans",
     "fleurs_tag":"cmn_hans_cn",
-    "commonvoice_hours":425.0,
+    "commonvoice_hours":426.0,
     "commonvoice_locale":"zh-TW",
     "in_benchmark":true
   },
@@ -43,7 +43,7 @@
     "family":"Indo-European",
     "flores_path":"spa_Latn",
     "fleurs_tag":"es_419",
-    "commonvoice_hours":451.0,
+    "commonvoice_hours":453.0,
     "commonvoice_locale":"es",
     "in_benchmark":true
   },
@@ -79,7 +79,7 @@
     "family":"Indo-European",
     "flores_path":"fra_Latn",
     "fleurs_tag":"fr_fr",
-    "commonvoice_hours":1097.0,
+    "commonvoice_hours":1104.0,
     "commonvoice_locale":"fr",
     "in_benchmark":true
   },
@@ -103,7 +103,7 @@
     "family":"Indo-European",
     "flores_path":"por_Latn",
     "fleurs_tag":"pt_br",
-    "commonvoice_hours":182.0,
+    "commonvoice_hours":183.0,
     "commonvoice_locale":"pt",
     "in_benchmark":true
   },
@@ -127,7 +127,7 @@
     "family":"Indo-European",
     "flores_path":"rus_Cyrl",
     "fleurs_tag":"ru_ru",
-    "commonvoice_hours":250.0,
+    "commonvoice_hours":252.0,
     "commonvoice_locale":"ru",
     "in_benchmark":true
   },
@@ -163,7 +163,7 @@
     "family":"Indo-European",
     "flores_path":"deu_Latn",
     "fleurs_tag":"de_de",
-    "commonvoice_hours":1390.0,
+    "commonvoice_hours":1394.0,
     "commonvoice_locale":"de",
     "in_benchmark":true
   },
@@ -187,7 +187,7 @@
     "family":"Dravidian",
     "flores_path":"tel_Telu",
     "fleurs_tag":"te_in",
-    "commonvoice_hours":0.6,
+    "commonvoice_hours":0.7,
     "commonvoice_locale":"te",
     "in_benchmark":true
   },
@@ -235,7 +235,7 @@
     "family":"Austroasiatic",
     "flores_path":"vie_Latn",
     "fleurs_tag":"vi_vn",
-    "commonvoice_hours":7.1,
+    "commonvoice_hours":7.6,
     "commonvoice_locale":"vi",
     "in_benchmark":true
   },
@@ -259,7 +259,7 @@
     "family":"Indo-European",
     "flores_path":"pes_Arab",
     "fleurs_tag":"fa_ir",
-    "commonvoice_hours":372.0,
+    "commonvoice_hours":373.0,
     "commonvoice_locale":"fa",
     "in_benchmark":true
   },
@@ -319,7 +319,7 @@
     "family":"Indo-European",
     "flores_path":"ita_Latn",
     "fleurs_tag":"it_it",
-    "commonvoice_hours":364.0,
+    "commonvoice_hours":365.0,
     "commonvoice_locale":"it",
     "in_benchmark":true
   },
@@ -379,7 +379,7 @@
     "family":"Indo-European",
     "flores_path":null,
     "fleurs_tag":"ps_af",
-    "commonvoice_hours":3087.0,
+    "commonvoice_hours":3257.0,
     "commonvoice_locale":"ps",
     "in_benchmark":false
   },
@@ -547,7 +547,7 @@
     "family":"Afro-Asiatic",
     "flores_path":"gaz_Latn",
     "fleurs_tag":"om_et",
-    "commonvoice_hours":0.0,
+    "commonvoice_hours":25.0,
     "commonvoice_locale":"om",
     "in_benchmark":true
   },
@@ -643,7 +643,7 @@
     "family":"Indo-European",
     "flores_path":"ukr_Cyrl",
     "fleurs_tag":"uk_ua",
-    "commonvoice_hours":100.0,
+    "commonvoice_hours":101.0,
     "commonvoice_locale":"uk",
     "in_benchmark":true
   },
@@ -655,7 +655,7 @@
     "family":"Atlantic-Congo",
     "flores_path":"yor_Latn",
     "fleurs_tag":"yo_ng",
-    "commonvoice_hours":6.4,
+    "commonvoice_hours":6.5,
     "commonvoice_locale":"yo",
     "in_benchmark":true
   },
@@ -679,7 +679,7 @@
     "family":"Atlantic-Congo",
     "flores_path":"ibo_Latn",
     "fleurs_tag":"ig_ng",
-    "commonvoice_hours":2.3,
+    "commonvoice_hours":2.5,
     "commonvoice_locale":"ig",
     "in_benchmark":true
   },
@@ -751,7 +751,7 @@
     "family":"Indo-European",
     "flores_path":"ron_Latn",
     "fleurs_tag":"ro_ro",
-    "commonvoice_hours":21.0,
+    "commonvoice_hours":24.0,
     "commonvoice_locale":"ro",
     "in_benchmark":true
   },
@@ -775,7 +775,7 @@
     "family":"Indo-European",
     "flores_path":"npi_Deva",
     "fleurs_tag":"ne_np",
-    "commonvoice_hours":1.3,
+    "commonvoice_hours":1.4,
     "commonvoice_locale":"ne-NP",
     "in_benchmark":true
   },
@@ -931,7 +931,7 @@
     "family":"Austroasiatic",
     "flores_path":"khm_Khmr",
     "fleurs_tag":"km_kh",
-    "commonvoice_hours":0.0,
+    "commonvoice_hours":0.2,
     "commonvoice_locale":"km",
     "in_benchmark":true
   },
@@ -1027,7 +1027,7 @@
     "family":"Uralic",
     "flores_path":"hun_Latn",
     "fleurs_tag":"hu_hu",
-    "commonvoice_hours":129.0,
+    "commonvoice_hours":136.0,
     "commonvoice_locale":"hu",
     "in_benchmark":true
   },
@@ -1075,7 +1075,7 @@
     "family":"Atlantic-Congo",
     "flores_path":"twi_Latn_akua1239",
     "fleurs_tag":null,
-    "commonvoice_hours":0.2,
+    "commonvoice_hours":0.3,
     "commonvoice_locale":"tw",
     "in_benchmark":true
   },
@@ -1195,7 +1195,7 @@
     "family":"Indo-European",
     "flores_path":"bel_Cyrl",
     "fleurs_tag":"be_by",
-    "commonvoice_hours":1819.0,
+    "commonvoice_hours":1822.0,
     "commonvoice_locale":"be",
     "in_benchmark":true
   },
@@ -1303,7 +1303,7 @@
     "family":"Indo-European",
     "flores_path":"cat_Latn",
     "fleurs_tag":"ca_es",
-    "commonvoice_hours":2945.0,
+    "commonvoice_hours":2983.0,
     "commonvoice_locale":"ca",
     "in_benchmark":true
   },
@@ -1387,7 +1387,7 @@
     "family":"Turkic",
     "flores_path":"uig_Arab",
     "fleurs_tag":null,
-    "commonvoice_hours":453.0,
+    "commonvoice_hours":462.0,
     "commonvoice_locale":"ug",
     "in_benchmark":true
   },
@@ -1411,7 +1411,7 @@
     "family":"Indo-European",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":0.7,
+    "commonvoice_hours":0.8,
     "commonvoice_locale":"gsw",
     "in_benchmark":false
   },
@@ -1435,7 +1435,7 @@
     "family":"Afro-Asiatic",
     "flores_path":"zgh_Tfng",
     "fleurs_tag":null,
-    "commonvoice_hours":1.4,
+    "commonvoice_hours":1.5,
     "commonvoice_locale":"zgh",
     "in_benchmark":true
   },
@@ -1555,7 +1555,7 @@
     "family":"Indo-European",
     "flores_path":"als_Latn",
     "fleurs_tag":null,
-    "commonvoice_hours":9.0,
+    "commonvoice_hours":9.1,
     "commonvoice_locale":"sq",
     "in_benchmark":true
   },
@@ -1711,7 +1711,7 @@
     "family":"Atlantic-Congo",
     "flores_path":"lug_Latn",
     "fleurs_tag":"lg_ug",
-    "commonvoice_hours":437.0,
+    "commonvoice_hours":438.0,
     "commonvoice_locale":"lg",
     "in_benchmark":true
   },
@@ -1771,7 +1771,7 @@
     "family":"Indo-European",
     "flores_path":"hye_Armn",
     "fleurs_tag":"hy_am",
-    "commonvoice_hours":34.0,
+    "commonvoice_hours":38.0,
     "commonvoice_locale":"hy-AM",
     "in_benchmark":true
   },
@@ -2143,7 +2143,7 @@
     "family":"Kartvelian",
     "flores_path":"kat_Geor",
     "fleurs_tag":"ka_ge",
-    "commonvoice_hours":168.0,
+    "commonvoice_hours":266.0,
     "commonvoice_locale":"ka",
     "in_benchmark":true
   },
@@ -2155,7 +2155,7 @@
     "family":"Indo-European",
     "flores_path":"glg_Latn",
     "fleurs_tag":"gl_es",
-    "commonvoice_hours":290.0,
+    "commonvoice_hours":324.0,
     "commonvoice_locale":"gl",
     "in_benchmark":true
   },
@@ -2215,8 +2215,8 @@
     "family":"Atlantic-Congo",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":null,
-    "commonvoice_locale":null,
+    "commonvoice_hours":0.0,
+    "commonvoice_locale":"tiv",
     "in_benchmark":false
   },
   {
@@ -2227,7 +2227,7 @@
     "family":"Afro-Asiatic",
     "flores_path":"kab_Latn",
     "fleurs_tag":null,
-    "commonvoice_hours":571.0,
+    "commonvoice_hours":572.0,
     "commonvoice_locale":"kab",
     "in_benchmark":true
   },
@@ -2371,8 +2371,8 @@
     "family":"Atlantic-Congo",
     "flores_path":"sag_Latn",
     "fleurs_tag":null,
-    "commonvoice_hours":null,
-    "commonvoice_locale":null,
+    "commonvoice_hours":0.0,
+    "commonvoice_locale":"sg",
     "in_benchmark":true
   },
   {
@@ -2503,7 +2503,7 @@
     "family":"Indo-European",
     "flores_path":"lit_Latn",
     "fleurs_tag":"lt_lt",
-    "commonvoice_hours":26.0,
+    "commonvoice_hours":27.0,
     "commonvoice_locale":"lt",
     "in_benchmark":true
   },
@@ -3007,7 +3007,7 @@
     "family":"Indo-European",
     "flores_path":"mkd_Cyrl",
     "fleurs_tag":"mk_mk",
-    "commonvoice_hours":23.0,
+    "commonvoice_hours":24.0,
     "commonvoice_locale":"mk",
     "in_benchmark":true
   },
@@ -3079,8 +3079,8 @@
     "family":"Atlantic-Congo",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":null,
-    "commonvoice_locale":null,
+    "commonvoice_hours":0.0,
+    "commonvoice_locale":"bin",
     "in_benchmark":false
   },
   {
@@ -3463,7 +3463,7 @@
     "family":"Indo-European",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":1.9,
+    "commonvoice_hours":2.0,
     "commonvoice_locale":"zza",
     "in_benchmark":false
   },
@@ -3475,7 +3475,7 @@
     "family":"Indo-European",
     "flores_path":"lvs_Latn",
     "fleurs_tag":"lv_lv",
-    "commonvoice_hours":266.0,
+    "commonvoice_hours":268.0,
     "commonvoice_locale":"lv",
     "in_benchmark":true
   },
@@ -3547,7 +3547,7 @@
     "family":"Abkhaz-Adyge",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":268.0,
+    "commonvoice_hours":271.0,
     "commonvoice_locale":"kbd",
     "in_benchmark":false
   },
@@ -3667,7 +3667,7 @@
     "family":"Indo-European",
     "flores_path":"ydd_Hebr",
     "fleurs_tag":null,
-    "commonvoice_hours":2.0,
+    "commonvoice_hours":2.1,
     "commonvoice_locale":"yi",
     "in_benchmark":true
   },
@@ -3763,8 +3763,8 @@
     "family":"Austroasiatic",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":null,
-    "commonvoice_locale":null,
+    "commonvoice_hours":0.0,
+    "commonvoice_locale":"mnw",
     "in_benchmark":false
   },
   {
@@ -4099,7 +4099,7 @@
     "family":"Indo-European",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":66.0,
+    "commonvoice_hours":67.0,
     "commonvoice_locale":"fy-NL",
     "in_benchmark":false
   },
@@ -4339,7 +4339,7 @@
     "family":"Indo-European",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":32.0,
+    "commonvoice_hours":33.0,
     "commonvoice_locale":"br",
     "in_benchmark":false
   },
@@ -4579,7 +4579,7 @@
     "family":"Afro-Asiatic",
     "flores_path":"mlt_Latn",
     "fleurs_tag":"mt_mt",
-    "commonvoice_hours":8.7,
+    "commonvoice_hours":8.8,
     "commonvoice_locale":"mt",
     "in_benchmark":true
   },
@@ -4639,7 +4639,7 @@
     "family":"Abkhaz-Adyge",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":71.0,
+    "commonvoice_hours":77.0,
     "commonvoice_locale":"ady",
     "in_benchmark":false
   },
@@ -4999,7 +4999,7 @@
     "family":"Nakh-Daghestanian",
     "flores_path":"dar_Cyrl",
     "fleurs_tag":null,
-    "commonvoice_hours":15.0,
+    "commonvoice_hours":16.0,
     "commonvoice_locale":"dar",
     "in_benchmark":true
   },
@@ -5227,8 +5227,8 @@
     "family":"Dravidian",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":null,
-    "commonvoice_locale":null,
+    "commonvoice_hours":0.0,
+    "commonvoice_locale":"bfq",
     "in_benchmark":false
   },
   {
@@ -5779,8 +5779,8 @@
     "family":"Atlantic-Congo",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":null,
-    "commonvoice_locale":null,
+    "commonvoice_hours":0.0,
+    "commonvoice_locale":"swb",
     "in_benchmark":false
   },
   {
@@ -6211,7 +6211,7 @@
     "family":"Abkhaz-Adyge",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":216.0,
+    "commonvoice_hours":217.0,
     "commonvoice_locale":"ab",
     "in_benchmark":false
   },
@@ -6871,7 +6871,7 @@
     "family":"Kartvelian",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":26.0,
+    "commonvoice_hours":44.0,
     "commonvoice_locale":"lzz",
     "in_benchmark":false
   },
@@ -7039,7 +7039,7 @@
     "family":"Indo-European",
     "flores_path":null,
     "fleurs_tag":null,
-    "commonvoice_hours":3.8,
+    "commonvoice_hours":3.9,
     "commonvoice_locale":"hsb",
     "in_benchmark":false
   },
@@ -7867,7 +7867,7 @@
     "family":"Artificial Language",
     "flores_path":"epo_Latn",
     "fleurs_tag":null,
-    "commonvoice_hours":1440.0,
+    "commonvoice_hours":1442.0,
     "commonvoice_locale":"eo",
     "in_benchmark":true
   },

results/model_health.json

@@ -1,4 +1,102 @@
 [
+  {
+    "model":"deepseek\/deepseek-v3.2-speciale-20251201",
+    "total":41,
+    "failed":41,
+    "score_nonfailed":null,
+    "failed_pct":100.0
+  },
+  {
+    "model":"baidu\/ernie-4.5-21b-a3b",
+    "total":41,
+    "failed":41,
+    "score_nonfailed":null,
+    "failed_pct":100.0
+  },
+  {
+    "model":"writer\/palmyra-x5-20250428",
+    "total":15030,
+    "failed":14820,
+    "score_nonfailed":0.5586882343,
+    "failed_pct":98.6027944112
+  },
+  {
+    "model":"qwen\/qwen3-max",
+    "total":15030,
+    "failed":14683,
+    "score_nonfailed":0.5747516232,
+    "failed_pct":97.6912840985
+  },
+  {
+    "model":"qwen\/qwen3.5-9b-20260310",
+    "total":15030,
+    "failed":13688,
+    "score_nonfailed":0.894407076,
+    "failed_pct":91.0711909514
+  },
+  {
+    "model":"nvidia\/llama-3.3-nemotron-super-49b-v1.5",
+    "total":15030,
+    "failed":13455,
+    "score_nonfailed":0.6289847358,
+    "failed_pct":89.5209580838
+  },
+  {
+    "model":"qwen\/qwen3.6-27b-20260422",
+    "total":15030,
+    "failed":12655,
+    "score_nonfailed":0.8270723406,
+    "failed_pct":84.1982701264
+  },
+  {
+    "model":"qwen\/qwen3.5-397b-a17b-20260216",
+    "total":15030,
+    "failed":12500,
+    "score_nonfailed":0.8238520798,
+    "failed_pct":83.1669993347
+  },
+  {
+    "model":"deepcogito\/cogito-v2.1-671b-20251118",
+    "total":15030,
+    "failed":12478,
+    "score_nonfailed":0.5737762836,
+    "failed_pct":83.0206254158
+  },
+  {
+    "model":"z-ai\/glm-4.7-flash-20260119",
+    "total":15030,
+    "failed":12165,
+    "score_nonfailed":0.7328015015,
+    "failed_pct":80.9381237525
+  },
+  {
+    "model":"z-ai\/glm-5-turbo-20260315",
+    "total":15030,
+    "failed":12145,
+    "score_nonfailed":0.8322476499,
+    "failed_pct":80.8050565536
+  },
+  {
+    "model":"nvidia\/nemotron-nano-9b-v2",
+    "total":15030,
+    "failed":12002,
+    "score_nonfailed":0.4370867053,
+    "failed_pct":79.8536260812
+  },
+  {
+    "model":"nvidia\/nemotron-3-super-120b-a12b-20230311",
+    "total":15030,
+    "failed":11274,
+    "score_nonfailed":0.3930247453,
+    "failed_pct":75.0099800399
+  },
+  {
+    "model":"nousresearch\/hermes-4-405b",
+    "total":15030,
+    "failed":10151,
+    "score_nonfailed":0.2702287054,
+    "failed_pct":67.5382568197
+  },
   {
     "model":"z-ai\/glm-4.6",
     "total":15030,
@@ -6,6 +104,69 @@
     "score_nonfailed":0.8472325073,
     "failed_pct":67.2588157019
   },
+  {
+    "model":"mistralai\/mistral-large-2512",
+    "total":15030,
+    "failed":9303,
+    "score_nonfailed":0.5400137224,
+    "failed_pct":61.8962075848
+  },
+  {
+    "model":"upstage\/solar-pro-3",
+    "total":15030,
+    "failed":8943,
+    "score_nonfailed":0.2903301643,
+    "failed_pct":59.500998004
+  },
+  {
+    "model":"google\/gemma-4-26b-a4b-it-20260403",
+    "total":15030,
+    "failed":8600,
+    "score_nonfailed":0.6435805501,
+    "failed_pct":57.2188955422
+  },
+  {
+    "model":"nvidia\/nemotron-3-nano-30b-a3b",
+    "total":15030,
+    "failed":8591,
+    "score_nonfailed":0.3274305545,
+    "failed_pct":57.1590153027
+  },
+  {
+    "model":"z-ai\/glm-5v-turbo-20260401",
+    "total":41,
+    "failed":23,
+    "score_nonfailed":0.9684492719,
+    "failed_pct":56.0975609756
+  },
+  {
+    "model":"mistralai\/mistral-small-2603",
+    "total":15030,
+    "failed":8335,
+    "score_nonfailed":0.7103749373,
+    "failed_pct":55.4557551564
+  },
+  {
+    "model":"qwen\/qwen3.5-122b-a10b-20260224",
+    "total":15030,
+    "failed":8218,
+    "score_nonfailed":0.6658006801,
+    "failed_pct":54.6773120426
+  },
+  {
+    "model":"deepseek\/deepseek-v3.1-terminus",
+    "total":15030,
+    "failed":8071,
+    "score_nonfailed":0.6103529696,
+    "failed_pct":53.6992681304
+  },
+  {
+    "model":"qwen\/qwen3.6-35b-a3b-20260415",
+    "total":15030,
+    "failed":7964,
+    "score_nonfailed":0.6591570315,
+    "failed_pct":52.9873586161
+  },
   {
     "model":"minimax\/minimax-m2.5",
     "total":15030,
@@ -13,6 +174,34 @@
     "score_nonfailed":0.716766251,
     "failed_pct":47.4384564205
   },
+  {
+    "model":"stepfun\/step-3.5-flash",
+    "total":41,
+    "failed":19,
+    "score_nonfailed":0.7455997951,
+    "failed_pct":46.3414634146
+  },
+  {
+    "model":"moonshotai\/kimi-k2.6",
+    "total":15080,
+    "failed":6828,
+    "score_nonfailed":0.714649433,
+    "failed_pct":45.2785145889
+  },
+  {
+    "model":"rekaai\/reka-edge-2603",
+    "total":15030,
+    "failed":6279,
+    "score_nonfailed":0.19356793,
+    "failed_pct":41.7764471058
+  },
+  {
+    "model":"stepfun\/step-3.7-flash-20260528",
+    "total":15030,
+    "failed":6198,
+    "score_nonfailed":0.7388377159,
+    "failed_pct":41.2375249501
+  },
   {
     "model":"qwen\/qwen3-235b-a22b",
     "total":15030,
@@ -20,6 +209,90 @@
     "score_nonfailed":0.5493210221,
     "failed_pct":40.4191616766
   },
+  {
+    "model":"ai21\/jamba-large-1.7",
+    "total":15080,
+    "failed":5789,
+    "score_nonfailed":0.4810803786,
+    "failed_pct":38.3885941645
+  },
+  {
+    "model":"prime-intellect\/intellect-3-20251126",
+    "total":15030,
+    "failed":5738,
+    "score_nonfailed":0.3868936552,
+    "failed_pct":38.1769793746
+  },
+  {
+    "model":"tencent\/hunyuan-a13b-instruct",
+    "total":15030,
+    "failed":5564,
+    "score_nonfailed":0.3154198804,
+    "failed_pct":37.0192947438
+  },
+  {
+    "model":"z-ai\/glm-4.5-air",
+    "total":15030,
+    "failed":4799,
+    "score_nonfailed":0.5064681541,
+    "failed_pct":31.9294743846
+  },
+  {
+    "model":"z-ai\/glm-4.5v",
+    "total":41,
+    "failed":13,
+    "score_nonfailed":0.8509373627,
+    "failed_pct":31.7073170732
+  },
+  {
+    "model":"arcee-ai\/trinity-mini-20251201",
+    "total":15080,
+    "failed":4602,
+    "score_nonfailed":0.3631567738,
+    "failed_pct":30.5172413793
+  },
+  {
+    "model":"ibm-granite\/granite-4.0-h-micro",
+    "total":15080,
+    "failed":4246,
+    "score_nonfailed":0.3396400581,
+    "failed_pct":28.1564986737
+  },
+  {
+    "model":"nousresearch\/hermes-4-70b",
+    "total":15080,
+    "failed":3832,
+    "score_nonfailed":0.2473206013,
+    "failed_pct":25.4111405836
+  },
+  {
+    "model":"amazon\/nova-2-lite-v1",
+    "total":15080,
+    "failed":3710,
+    "score_nonfailed":0.535897764,
+    "failed_pct":24.6021220159
+  },
+  {
+    "model":"deepseek\/deepseek-chat-v3.1",
+    "total":15080,
+    "failed":3299,
+    "score_nonfailed":0.5682361214,
+    "failed_pct":21.8766578249
+  },
+  {
+    "model":"liquid\/lfm-2-24b-a2b-20260224",
+    "total":15080,
+    "failed":2808,
+    "score_nonfailed":0.349189204,
+    "failed_pct":18.6206896552
+  },
+  {
+    "model":"bytedance-seed\/seed-1.6-flash-20250625",
+    "total":15080,
+    "failed":2783,
+    "score_nonfailed":0.3902097008,
+    "failed_pct":18.4549071618
+  },
   {
     "model":"google\/translate-v2",
     "total":7760,
@@ -27,19 +300,68 @@
     "score_nonfailed":0.4370955661,
     "failed_pct":18.0412371134
   },
+  {
+    "model":"deepseek\/deepseek-r1-0528",
+    "total":41,
+    "failed":7,
+    "score_nonfailed":0.816956768,
+    "failed_pct":17.0731707317
+  },
+  {
+    "model":"openai\/gpt-oss-20b",
+    "total":15030,
+    "failed":2547,
+    "score_nonfailed":0.4648651834,
+    "failed_pct":16.9461077844
+  },
+  {
+    "model":"microsoft\/phi-4-mini-instruct",
+    "total":15080,
+    "failed":2517,
+    "score_nonfailed":0.2620524844,
+    "failed_pct":16.6909814324
+  },
+  {
+    "model":"google\/gemma-4-31b-it-20260402",
+    "total":15080,
+    "failed":2507,
+    "score_nonfailed":0.6070484846,
+    "failed_pct":16.624668435
+  },
+  {
+    "model":"z-ai\/glm-4-32b-0414",
+    "total":15030,
+    "failed":2492,
+    "score_nonfailed":0.4939664853,
+    "failed_pct":16.5801729874
+  },
+  {
+    "model":"google\/gemini-3.1-flash-lite",
+    "total":15080,
+    "failed":1827,
+    "score_nonfailed":0.6534476235,
+    "failed_pct":12.1153846154
+  },
   {
     "model":"qwen\/qwen3-30b-a3b",
     "total":15030,
-    "failed":2186,
-    "score_nonfailed":0.4860680279,
-    "failed_pct":14.5442448436
+    "failed":1813,
+    "score_nonfailed":0.4806886966,
+    "failed_pct":12.0625415835
   },
   {
-    "model":"anthropic\/claude-sonnet-4.6",
+    "model":"nvidia\/nemotron-3-ultra-550b-a55b-20260604",
+    "total":15080,
+    "failed":1633,
+    "score_nonfailed":0.5317395427,
+    "failed_pct":10.8289124668
+  },
+  {
+    "model":"qwen\/qwen3-30b-a3b-instruct-2507",
     "total":15030,
-    "failed":1193,
-    "score_nonfailed":0.6553656376,
-    "failed_pct":7.9374584165
+    "failed":1252,
+    "score_nonfailed":0.4588521485,
+    "failed_pct":8.3300066534
   },
   {
     "model":"x-ai\/grok-4.1-fast",
@@ -49,11 +371,25 @@
     "failed_pct":7.9308050566
   },
   {
-    "model":"openai\/gpt-5",
+    "model":"qwen\/qwen3-next-80b-a3b-instruct-2509",
     "total":15030,
-    "failed":1175,
-    "score_nonfailed":0.6563698813,
-    "failed_pct":7.8176979375
+    "failed":1146,
+    "score_nonfailed":0.518827058,
+    "failed_pct":7.624750499
+  },
+  {
+    "model":"google\/gemma-3n-e4b-it",
+    "total":41,
+    "failed":3,
+    "score_nonfailed":0.7847097861,
+    "failed_pct":7.3170731707
+  },
+  {
+    "model":"openai\/gpt-5",
+    "total":15080,
+    "failed":1062,
+    "score_nonfailed":0.6516557567,
+    "failed_pct":7.0424403183
   },
   {
     "model":"allenai\/olmo-3.1-32b-instruct",
@@ -63,11 +399,53 @@
     "failed_pct":6.3007318696
   },
   {
-    "model":"deepseek\/deepseek-v3.2-exp",
+    "model":"mistralai\/ministral-3b-2512",
+    "total":15080,
+    "failed":925,
+    "score_nonfailed":0.3631905149,
+    "failed_pct":6.1339522546
+  },
+  {
+    "model":"anthropic\/claude-sonnet-4.6",
+    "total":15080,
+    "failed":873,
+    "score_nonfailed":0.6504398418,
+    "failed_pct":5.7891246684
+  },
+  {
+    "model":"mistralai\/ministral-8b-2512",
+    "total":15080,
+    "failed":709,
+    "score_nonfailed":0.4329534856,
+    "failed_pct":4.7015915119
+  },
+  {
+    "model":"deepseek\/deepseek-v4-flash-20260423",
+    "total":15080,
+    "failed":616,
+    "score_nonfailed":0.6238057314,
+    "failed_pct":4.0848806366
+  },
+  {
+    "model":"mistralai\/ministral-14b-2512",
+    "total":15080,
+    "failed":612,
+    "score_nonfailed":0.4586929167,
+    "failed_pct":4.0583554377
+  },
+  {
+    "model":"qwen\/qwen3.6-flash",
     "total":15030,
-    "failed":920,
-    "score_nonfailed":0.5855771228,
-    "failed_pct":6.121091151
+    "failed":554,
+    "score_nonfailed":0.5823627104,
+    "failed_pct":3.6859614105
+  },
+  {
+    "model":"ibm-granite\/granite-4.1-8b-20260429",
+    "total":15080,
+    "failed":553,
+    "score_nonfailed":0.4250007907,
+    "failed_pct":3.6671087533
   },
   {
     "model":"anthropic\/claude-opus-4.6",
@@ -76,47 +454,124 @@
     "score_nonfailed":0.6793265387,
     "failed_pct":3.5662009315
   },
+  {
+    "model":"deepseek\/deepseek-v3.2-exp",
+    "total":15080,
+    "failed":504,
+    "score_nonfailed":0.5810977119,
+    "failed_pct":3.3421750663
+  },
+  {
+    "model":"openai\/gpt-5-nano",
+    "total":15080,
+    "failed":358,
+    "score_nonfailed":0.5014027303,
+    "failed_pct":2.374005305
+  },
+  {
+    "model":"bytedance-seed\/seed-2.0-mini-20260224",
+    "total":15080,
+    "failed":189,
+    "score_nonfailed":0.5152390813,
+    "failed_pct":1.2533156499
+  },
   {
     "model":"google\/gemini-3.1-pro-preview",
+    "total":15080,
+    "failed":188,
+    "score_nonfailed":0.6897643923,
+    "failed_pct":1.2466843501
+  },
+  {
+    "model":"bytedance-seed\/seed-1.6-20250625",
+    "total":15080,
+    "failed":177,
+    "score_nonfailed":0.5004845457,
+    "failed_pct":1.1737400531
+  },
+  {
+    "model":"openai\/gpt-5.5",
     "total":15030,
-    "failed":534,
-    "score_nonfailed":0.6915737079,
-    "failed_pct":3.5528942116
+    "failed":168,
+    "score_nonfailed":0.6520055238,
+    "failed_pct":1.1177644711
   },
   {
-    "model":"openai\/gpt-5-nano",
+    "model":"qwen\/qwen3.6-plus-04-02",
     "total":15030,
-    "failed":453,
-    "score_nonfailed":0.5038289898,
-    "failed_pct":3.0139720559
+    "failed":155,
+    "score_nonfailed":0.6258724148,
+    "failed_pct":1.0312707917
   },
   {
     "model":"openai\/gpt-5.4",
     "total":15030,
-    "failed":247,
-    "score_nonfailed":0.6320077882,
-    "failed_pct":1.6433799069
+    "failed":150,
+    "score_nonfailed":0.6309648172,
+    "failed_pct":0.998003992
   },
   {
-    "model":"openai\/gpt-5.2",
+    "model":"mistralai\/mistral-small-3.2-24b-instruct",
+    "total":15080,
+    "failed":139,
+    "score_nonfailed":0.4631506534,
+    "failed_pct":0.9217506631
+  },
+  {
+    "model":"anthropic\/claude-opus-4.8",
+    "total":15080,
+    "failed":128,
+    "score_nonfailed":0.661031115,
+    "failed_pct":0.848806366
+  },
+  {
+    "model":"openai\/gpt-5.3-chat-20260303",
     "total":15030,
-    "failed":96,
-    "score_nonfailed":0.6242762659,
-    "failed_pct":0.6387225549
+    "failed":109,
+    "score_nonfailed":0.6358423209,
+    "failed_pct":0.7252162342
   },
   {
-    "model":"openai\/gpt-5-mini",
+    "model":"anthropic\/claude-opus-4.7",
+    "total":15080,
+    "failed":108,
+    "score_nonfailed":0.6685617849,
+    "failed_pct":0.7161803714
+  },
+  {
+    "model":"x-ai\/grok-4.20",
     "total":15030,
-    "failed":81,
-    "score_nonfailed":0.5571339927,
-    "failed_pct":0.5389221557
+    "failed":89,
+    "score_nonfailed":0.6311292606,
+    "failed_pct":0.5921490353
+  },
+  {
+    "model":"openai\/gpt-5.2",
+    "total":15080,
+    "failed":86,
+    "score_nonfailed":0.6231101124,
+    "failed_pct":0.5702917772
+  },
+  {
+    "model":"bytedance-seed\/seed-2.0-lite-20260309",
+    "total":15080,
+    "failed":75,
+    "score_nonfailed":0.5853661867,
+    "failed_pct":0.4973474801
+  },
+  {
+    "model":"openai\/gpt-5-mini",
+    "total":15080,
+    "failed":66,
+    "score_nonfailed":0.5558036623,
+    "failed_pct":0.4376657825
   },
   {
     "model":"openai\/gpt-5.1",
-    "total":15030,
-    "failed":35,
-    "score_nonfailed":0.6182095927,
-    "failed_pct":0.2328675981
+    "total":15080,
+    "failed":28,
+    "score_nonfailed":0.6170952803,
+    "failed_pct":0.1856763926
   },
   {
     "model":"anthropic\/claude-opus-4.5",
@@ -127,247 +582,247 @@
   },
   {
     "model":"anthropic\/claude-sonnet-4.5",
-    "total":15030,
-    "failed":19,
-    "score_nonfailed":0.644467649,
-    "failed_pct":0.126413839
+    "total":15080,
+    "failed":17,
+    "score_nonfailed":0.6437023017,
+    "failed_pct":0.1127320955
   },
   {
     "model":"amazon\/nova-pro-v1",
-    "total":15030,
-    "failed":8,
-    "score_nonfailed":0.5459768526,
-    "failed_pct":0.0532268796
+    "total":15080,
+    "failed":7,
+    "score_nonfailed":0.5452692078,
+    "failed_pct":0.0464190981
   },
   {
     "model":"openai\/gpt-oss-120b",
     "total":15030,
-    "failed":8,
-    "score_nonfailed":0.5131642507,
-    "failed_pct":0.0532268796
+    "failed":4,
+    "score_nonfailed":0.5130942947,
+    "failed_pct":0.0266134398
   },
   {
-    "model":"mistralai\/mistral-nemo",
-    "total":15030,
-    "failed":1,
-    "score_nonfailed":0.3401549906,
-    "failed_pct":0.0066533599
-  },
-  {
-    "model":"qwen\/qwen3-vl-30b-a3b-thinking",
+    "model":"perplexity\/sonar-reasoning-pro",
     "total":17,
     "failed":0,
-    "score_nonfailed":0.5794774559,
+    "score_nonfailed":0.7141204358,
     "failed_pct":0.0
   },
   {
-    "model":"anthropic\/claude-3-7-sonnet-20250219",
-    "total":2818,
+    "model":"alpindale\/goliath-120b",
+    "total":170,
     "failed":0,
-    "score_nonfailed":0.6793719385,
+    "score_nonfailed":0.3700368906,
     "failed_pct":0.0
   },
   {
-    "model":"openai\/gpt-3.5-turbo-0613",
-    "total":170,
+    "model":"amazon\/nova-micro-v1",
+    "total":17,
     "failed":0,
-    "score_nonfailed":0.5504296912,
+    "score_nonfailed":0.6700323243,
     "failed_pct":0.0
   },
   {
-    "model":"z-ai\/glm-4.5-air",
-    "total":140,
+    "model":"openai\/o4-mini-2025-04-16",
+    "total":167,
     "failed":0,
-    "score_nonfailed":0.582694482,
+    "score_nonfailed":0.646647725,
     "failed_pct":0.0
   },
   {
-    "model":"openai\/gpt-4.1",
-    "total":15030,
+    "model":"x-ai\/grok-3",
+    "total":170,
     "failed":0,
-    "score_nonfailed":0.5585604761,
+    "score_nonfailed":0.6330974229,
     "failed_pct":0.0
   },
   {
-    "model":"openai\/gpt-4.1-mini",
-    "total":170,
+    "model":"x-ai\/grok-4-fast",
+    "total":15030,
     "failed":0,
-    "score_nonfailed":0.6424898144,
+    "score_nonfailed":0.5469999688,
     "failed_pct":0.0
   },
   {
-    "model":"openai\/gpt-4.1-nano",
-    "total":170,
+    "model":"x-ai\/grok-4",
+    "total":3475,
     "failed":0,
-    "score_nonfailed":0.6064425983,
+    "score_nonfailed":0.6712118876,
     "failed_pct":0.0
   },
   {
-    "model":"openai\/gpt-4o",
-    "total":15030,
+    "model":"qwen\/qwen3-8b-04-28",
+    "total":113,
     "failed":0,
-    "score_nonfailed":0.5881047852,
+    "score_nonfailed":0.6392222245,
     "failed_pct":0.0
   },
   {
-    "model":"openai\/gpt-4o-2024-11-20",
-    "total":17,
+    "model":"amazon\/nova-premier-v1",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.742557214,
+    "score_nonfailed":0.5266995001,
     "failed_pct":0.0
   },
   {
-    "model":"openai\/gpt-4o-mini",
-    "total":170,
+    "model":"anthropic\/claude-sonnet-4",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.6541716124,
+    "score_nonfailed":0.6235642707,
     "failed_pct":0.0
   },
   {
-    "model":"openai\/gpt-5-codex",
-    "total":170,
+    "model":"anthropic\/claude-3-7-sonnet-20250219",
+    "total":2818,
     "failed":0,
-    "score_nonfailed":0.7040281767,
+    "score_nonfailed":0.6793719385,
     "failed_pct":0.0
   },
   {
-    "model":"openai\/gpt-5-chat-2025-08-07",
-    "total":170,
+    "model":"anthropic\/claude-3.5-sonnet",
+    "total":14288,
     "failed":0,
-    "score_nonfailed":0.639942047,
+    "score_nonfailed":0.6279979071,
     "failed_pct":0.0
   },
   {
-    "model":"qwen\/qwen3-8b-04-28",
-    "total":113,
+    "model":"anthropic\/claude-3.7-sonnet",
+    "total":15030,
     "failed":0,
-    "score_nonfailed":0.6392222245,
+    "score_nonfailed":0.632849603,
     "failed_pct":0.0
   },
   {
-    "model":"amazon\/nova-micro-v1",
+    "model":"qwen\/qwen3-vl-30b-a3b-thinking",
     "total":17,
     "failed":0,
-    "score_nonfailed":0.6700323243,
+    "score_nonfailed":0.5794774559,
     "failed_pct":0.0
   },
   {
-    "model":"x-ai\/grok-4-fast",
-    "total":15030,
+    "model":"anthropic\/claude-haiku-4.5",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.5469999688,
+    "score_nonfailed":0.5380086427,
     "failed_pct":0.0
   },
   {
-    "model":"x-ai\/grok-4",
-    "total":3475,
+    "model":"openai\/gpt-3.5-turbo-0613",
+    "total":170,
     "failed":0,
-    "score_nonfailed":0.6712118876,
+    "score_nonfailed":0.5504296912,
     "failed_pct":0.0
   },
   {
-    "model":"nousresearch\/hermes-3-llama-3.1-405b",
+    "model":"openai\/gpt-5-codex",
     "total":170,
     "failed":0,
-    "score_nonfailed":0.6629729711,
+    "score_nonfailed":0.7040281767,
     "failed_pct":0.0
   },
   {
-    "model":"x-ai\/grok-3",
+    "model":"openai\/gpt-5-chat-2025-08-07",
     "total":170,
     "failed":0,
-    "score_nonfailed":0.6330974229,
+    "score_nonfailed":0.639942047,
     "failed_pct":0.0
   },
   {
-    "model":"openai\/o4-mini-2025-04-16",
-    "total":167,
+    "model":"microsoft\/phi-4-multimodal-instruct",
+    "total":17,
     "failed":0,
-    "score_nonfailed":0.646647725,
+    "score_nonfailed":0.7358366016,
     "failed_pct":0.0
   },
   {
-    "model":"perplexity\/sonar-reasoning-pro",
-    "total":17,
+    "model":"microsoft\/phi-4",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.7141204358,
+    "score_nonfailed":0.3449970743,
     "failed_pct":0.0
   },
   {
-    "model":"amazon\/nova-premier-v1",
-    "total":15030,
+    "model":"microsoft\/phi-3-mini-128k-instruct",
+    "total":170,
     "failed":0,
-    "score_nonfailed":0.5273180978,
+    "score_nonfailed":0.3182626307,
     "failed_pct":0.0
   },
   {
-    "model":"openai\/gpt-3.5-turbo",
-    "total":15030,
+    "model":"meta-llama\/llama-4-maverick",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.3505272925,
+    "score_nonfailed":0.5580584586,
     "failed_pct":0.0
   },
   {
-    "model":"mistralai\/mixtral-8x7b-instruct",
-    "total":170,
+    "model":"meta-llama\/llama-3.3-70b-instruct",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.4240249599,
+    "score_nonfailed":0.484181525,
     "failed_pct":0.0
   },
   {
-    "model":"neversleep\/noromaid-20b",
-    "total":170,
+    "model":"meta-llama\/llama-3.1-70b-instruct",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.2124753832,
+    "score_nonfailed":0.4854957629,
     "failed_pct":0.0
   },
   {
-    "model":"google\/gemma-3-27b-it",
-    "total":15030,
+    "model":"meta-llama\/llama-3-70b-instruct",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.5467921901,
+    "score_nonfailed":0.467426787,
     "failed_pct":0.0
   },
   {
-    "model":"anthropic\/claude-sonnet-4",
-    "total":15030,
+    "model":"cohere\/command-r-plus-08-2024",
+    "total":6369,
     "failed":0,
-    "score_nonfailed":0.6243490094,
+    "score_nonfailed":0.3479812964,
     "failed_pct":0.0
   },
   {
-    "model":"anthropic\/claude-haiku-4.5",
-    "total":15030,
+    "model":"google\/gemma-3-27b-it",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.5391267264,
+    "score_nonfailed":0.5460647648,
     "failed_pct":0.0
   },
   {
-    "model":"baidu\/ernie-4.5-300b-a47b",
+    "model":"google\/gemini-3-pro-preview",
     "total":15030,
     "failed":0,
-    "score_nonfailed":0.5751447662,
+    "score_nonfailed":0.6580387071,
     "failed_pct":0.0
   },
   {
-    "model":"cohere\/command-a",
-    "total":15030,
+    "model":"google\/gemini-2.5-pro",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.5053932299,
+    "score_nonfailed":0.6410620763,
     "failed_pct":0.0
   },
   {
-    "model":"cohere\/command-r-08-2024",
+    "model":"google\/gemini-2.5-flash-lite",
+    "total":15080,
+    "failed":0,
+    "score_nonfailed":0.5588659385,
+    "failed_pct":0.0
+  },
+  {
+    "model":"google\/gemini-2.5-flash-image",
     "total":170,
     "failed":0,
-    "score_nonfailed":0.5222153257,
+    "score_nonfailed":0.713401524,
     "failed_pct":0.0
   },
   {
-    "model":"cohere\/command-r-plus-08-2024",
-    "total":6369,
+    "model":"google\/gemini-2.5-flash",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.3479812964,
+    "score_nonfailed":0.4238758246,
     "failed_pct":0.0
   },
   {
@@ -378,129 +833,129 @@
     "failed_pct":0.0
   },
   {
-    "model":"google\/gemini-2.5-flash",
-    "total":15030,
+    "model":"cohere\/command-r-08-2024",
+    "total":170,
     "failed":0,
-    "score_nonfailed":0.4237452366,
+    "score_nonfailed":0.5222153257,
     "failed_pct":0.0
   },
   {
-    "model":"google\/gemini-2.5-flash-image",
-    "total":170,
+    "model":"mistralai\/magistral-small-2506",
+    "total":120,
     "failed":0,
-    "score_nonfailed":0.713401524,
+    "score_nonfailed":0.3314969313,
     "failed_pct":0.0
   },
   {
-    "model":"google\/gemini-2.5-flash-lite",
-    "total":15030,
+    "model":"cohere\/command-a",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.5594649326,
+    "score_nonfailed":0.5047155086,
     "failed_pct":0.0
   },
   {
-    "model":"google\/gemini-2.5-pro",
-    "total":15030,
+    "model":"openai\/gpt-3.5-turbo",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.6415432467,
+    "score_nonfailed":0.3503358897,
     "failed_pct":0.0
   },
   {
-    "model":"google\/gemini-3-pro-preview",
-    "total":15030,
+    "model":"openai\/gpt-4o-mini",
+    "total":170,
     "failed":0,
-    "score_nonfailed":0.6580387071,
+    "score_nonfailed":0.6541716124,
     "failed_pct":0.0
   },
   {
-    "model":"anthropic\/claude-3.7-sonnet",
-    "total":15030,
+    "model":"openai\/gpt-4o-2024-11-20",
+    "total":17,
     "failed":0,
-    "score_nonfailed":0.632849603,
+    "score_nonfailed":0.742557214,
     "failed_pct":0.0
   },
   {
-    "model":"moonshotai\/kimi-k2",
-    "total":15030,
+    "model":"openai\/gpt-4o",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.55318705,
+    "score_nonfailed":0.5872301625,
     "failed_pct":0.0
   },
   {
-    "model":"meta-llama\/llama-3-70b-instruct",
-    "total":15030,
+    "model":"openai\/gpt-4.1-nano",
+    "total":170,
     "failed":0,
-    "score_nonfailed":0.468044074,
+    "score_nonfailed":0.6064425983,
     "failed_pct":0.0
   },
   {
-    "model":"meta-llama\/llama-3.1-70b-instruct",
-    "total":15030,
+    "model":"openai\/gpt-4.1-mini",
+    "total":170,
     "failed":0,
-    "score_nonfailed":0.4860431126,
+    "score_nonfailed":0.6424898144,
     "failed_pct":0.0
   },
   {
-    "model":"meta-llama\/llama-3.3-70b-instruct",
-    "total":15030,
+    "model":"openai\/gpt-4.1",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.4846053463,
+    "score_nonfailed":0.5577101349,
     "failed_pct":0.0
   },
   {
-    "model":"meta-llama\/llama-4-maverick",
+    "model":"baidu\/ernie-4.5-300b-a47b",
     "total":15030,
     "failed":0,
-    "score_nonfailed":0.5585725974,
+    "score_nonfailed":0.5751447662,
     "failed_pct":0.0
   },
   {
-    "model":"microsoft\/phi-3-mini-128k-instruct",
-    "total":170,
+    "model":"mistralai\/mistral-medium-3.1",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.3182626307,
+    "score_nonfailed":0.5110552261,
     "failed_pct":0.0
   },
   {
-    "model":"microsoft\/phi-4",
-    "total":15030,
+    "model":"nousresearch\/hermes-3-llama-3.1-405b",
+    "total":170,
     "failed":0,
-    "score_nonfailed":0.3454741484,
+    "score_nonfailed":0.6629729711,
     "failed_pct":0.0
   },
   {
-    "model":"microsoft\/phi-4-multimodal-instruct",
-    "total":17,
+    "model":"neversleep\/noromaid-20b",
+    "total":170,
     "failed":0,
-    "score_nonfailed":0.7358366016,
+    "score_nonfailed":0.2124753832,
     "failed_pct":0.0
   },
   {
-    "model":"anthropic\/claude-3.5-sonnet",
-    "total":14288,
+    "model":"moonshotai\/kimi-k2",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.6279979071,
+    "score_nonfailed":0.5523425103,
     "failed_pct":0.0
   },
   {
-    "model":"mistralai\/magistral-small-2506",
-    "total":120,
+    "model":"mistralai\/mixtral-8x7b-instruct",
+    "total":170,
     "failed":0,
-    "score_nonfailed":0.3314969313,
+    "score_nonfailed":0.4240249599,
     "failed_pct":0.0
   },
   {
-    "model":"alpindale\/goliath-120b",
-    "total":170,
+    "model":"mistralai\/mistral-saba",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.3700368906,
+    "score_nonfailed":0.4559403599,
     "failed_pct":0.0
   },
   {
-    "model":"mistralai\/mistral-saba",
-    "total":15030,
+    "model":"mistralai\/mistral-nemo",
+    "total":15080,
     "failed":0,
-    "score_nonfailed":0.4565603305,
+    "score_nonfailed":0.3398077816,
     "failed_pct":0.0
   },
   {
@@ -509,12 +964,5 @@
     "failed":0,
     "score_nonfailed":0.70419565,
     "failed_pct":0.0
-  },
-  {
-    "model":"mistralai\/mistral-medium-3.1",
-    "total":15030,
-    "failed":0,
-    "score_nonfailed":0.5117549741,
-    "failed_pct":0.0
   }
 ]

results/models.json

@@ -1,15 +1,15 @@
 [
   {
-    "id":"allenai\/olmo-3.1-32b-instruct",
-    "name":"Olmo 3.1 32B Instruct",
-    "provider_name":"AllenAI",
-    "cost":0.6,
+    "id":"ai21\/jamba-large-1.7",
+    "name":"Jamba Large 1.7",
+    "provider_name":"AI21",
+    "cost":8.0,
     "train_on_prompts":false,
-    "hf_id":"allenai\/Olmo-3.1-32B-Instruct",
-    "size":null,
+    "hf_id":"ai21labs\/AI21-Jamba-Large-1.7",
+    "size":398555145696.0,
     "type":"open-source",
-    "license":"Apache 2.0",
-    "creation_date":1765324800000.0,
+    "license":"Other",
+    "creation_date":1751414400000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -21,16 +21,16 @@
     ]
   },
   {
-    "id":"amazon\/nova-premier-v1",
-    "name":"Nova Premier 1.0",
+    "id":"amazon\/nova-2-lite-v1",
+    "name":"Nova 2 Lite",
     "provider_name":"Amazon",
-    "cost":12.5,
+    "cost":2.5,
     "train_on_prompts":false,
     "hf_id":null,
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1761868800000.0,
+    "creation_date":1764633600000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -42,16 +42,16 @@
     ]
   },
   {
-    "id":"amazon\/nova-pro-v1",
-    "name":"Nova Pro 1.0",
+    "id":"amazon\/nova-premier-v1",
+    "name":"Nova Premier 1.0",
     "provider_name":"Amazon",
-    "cost":3.2,
+    "cost":12.5,
     "train_on_prompts":false,
     "hf_id":null,
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1733356800000.0,
+    "creation_date":1761868800000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -63,16 +63,16 @@
     ]
   },
   {
-    "id":"anthropic\/claude-3.7-sonnet",
-    "name":"Claude 3.7 Sonnet",
-    "provider_name":"Anthropic",
-    "cost":15.0,
+    "id":"amazon\/nova-pro-v1",
+    "name":"Nova Pro 1.0",
+    "provider_name":"Amazon",
+    "cost":3.2,
     "train_on_prompts":false,
     "hf_id":null,
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1740355200000.0,
+    "creation_date":1733356800000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -93,7 +93,7 @@
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1760486400000.0,
+    "creation_date":1760486400000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -105,8 +105,8 @@
     ]
   },
   {
-    "id":"anthropic\/claude-opus-4.5",
-    "name":"Claude Opus 4.5",
+    "id":"anthropic\/claude-opus-4.7",
+    "name":"Claude Opus 4.7",
     "provider_name":"Anthropic",
     "cost":25.0,
     "train_on_prompts":false,
@@ -114,7 +114,7 @@
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1763942400000.0,
+    "creation_date":1776297600000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -126,8 +126,8 @@
     ]
   },
   {
-    "id":"anthropic\/claude-opus-4.6",
-    "name":"Claude Opus 4.6",
+    "id":"anthropic\/claude-opus-4.8",
+    "name":"Claude Opus 4.8",
     "provider_name":"Anthropic",
     "cost":25.0,
     "train_on_prompts":false,
@@ -135,7 +135,7 @@
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1770163200000.0,
+    "creation_date":1779840000000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -156,7 +156,7 @@
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1747872000000.0,
+    "creation_date":1747872000000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -177,7 +177,7 @@
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1759104000000.0,
+    "creation_date":1759104000000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -198,28 +198,7 @@
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1771286400000.0,
-    "tasks":[
-      "translation_from",
-      "translation_to",
-      "classification",
-      "mmlu",
-      "arc",
-      "truthfulqa",
-      "mgsm"
-    ]
-  },
-  {
-    "id":"baidu\/ernie-4.5-300b-a47b",
-    "name":"ERNIE 4.5 300B A47B ",
-    "provider_name":"Baidu",
-    "cost":1.1,
-    "train_on_prompts":false,
-    "hf_id":"baidu\/ERNIE-4.5-300B-A47B-PT",
-    "size":null,
-    "type":"open-source",
-    "license":"Apache 2.0",
-    "creation_date":1751068800000.0,
+    "creation_date":1771286400000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -231,16 +210,16 @@
     ]
   },
   {
-    "id":"cohere\/command-a",
-    "name":"Command A",
-    "provider_name":"Cohere",
-    "cost":10.0,
+    "id":"arcee-ai\/trinity-mini-20251201",
+    "name":"Trinity Mini",
+    "provider_name":"Arcee AI",
+    "cost":0.15,
     "train_on_prompts":false,
-    "hf_id":"CohereLabs\/c4ai-command-a-03-2025",
-    "size":null,
+    "hf_id":"arcee-ai\/Trinity-Mini",
+    "size":26123974400.0,
     "type":"open-source",
-    "license":"Cc By Nc 4.0",
-    "creation_date":1741651200000.0,
+    "license":"Other",
+    "creation_date":1764547200000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -252,16 +231,16 @@
     ]
   },
   {
-    "id":"deepseek\/deepseek-v3.2-exp",
-    "name":"DeepSeek V3.2 Exp",
-    "provider_name":"DeepSeek",
-    "cost":0.41,
+    "id":"bytedance-seed\/seed-1.6-20250625",
+    "name":"Seed 1.6",
+    "provider_name":"ByteDance Seed",
+    "cost":2.0,
     "train_on_prompts":false,
-    "hf_id":"deepseek-ai\/DeepSeek-V3.2-Exp",
+    "hf_id":null,
     "size":null,
-    "type":"open-source",
-    "license":"Mit",
-    "creation_date":1759104000000.0,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1766448000000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -273,16 +252,16 @@
     ]
   },
   {
-    "id":"google\/gemini-2.5-flash",
-    "name":"Gemini 2.5 Flash",
-    "provider_name":"Google",
-    "cost":2.5,
+    "id":"bytedance-seed\/seed-1.6-flash-20250625",
+    "name":"Seed 1.6 Flash",
+    "provider_name":"ByteDance Seed",
+    "cost":0.3,
     "train_on_prompts":false,
     "hf_id":null,
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1750118400000.0,
+    "creation_date":1766448000000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -294,16 +273,16 @@
     ]
   },
   {
-    "id":"google\/gemini-2.5-flash-lite",
-    "name":"Gemini 2.5 Flash Lite",
-    "provider_name":"Google",
-    "cost":0.4,
+    "id":"bytedance-seed\/seed-2.0-lite-20260309",
+    "name":"Seed-2.0-Lite",
+    "provider_name":"ByteDance Seed",
+    "cost":2.0,
     "train_on_prompts":false,
     "hf_id":null,
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1753142400000.0,
+    "creation_date":1773100800000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -315,16 +294,16 @@
     ]
   },
   {
-    "id":"google\/gemini-2.5-pro",
-    "name":"Gemini 2.5 Pro",
-    "provider_name":"Google",
-    "cost":10.0,
+    "id":"bytedance-seed\/seed-2.0-mini-20260224",
+    "name":"Seed-2.0-Mini",
+    "provider_name":"ByteDance Seed",
+    "cost":0.4,
     "train_on_prompts":false,
     "hf_id":null,
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1750118400000.0,
+    "creation_date":1772064000000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -336,16 +315,16 @@
     ]
   },
   {
-    "id":"google\/gemini-3-pro-preview",
-    "name":"Gemini 3 Pro Preview",
-    "provider_name":"Google",
-    "cost":12.0,
+    "id":"cohere\/command-a",
+    "name":"Command A",
+    "provider_name":"Cohere",
+    "cost":10.0,
     "train_on_prompts":false,
-    "hf_id":null,
-    "size":null,
-    "type":"closed-source",
-    "license":null,
-    "creation_date":1763424000000.0,
+    "hf_id":"CohereLabs\/c4ai-command-a-03-2025",
+    "size":111057580032.0,
+    "type":"open-source",
+    "license":"Cc By Nc 4.0",
+    "creation_date":1741651200000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -357,16 +336,16 @@
     ]
   },
   {
-    "id":"google\/gemini-3.1-pro-preview",
-    "name":"Gemini 3.1 Pro Preview",
-    "provider_name":"Google",
-    "cost":12.0,
+    "id":"deepcogito\/cogito-v2.1-671b-20251118",
+    "name":"Cogito v2.1 671B",
+    "provider_name":"Deep Cogito",
+    "cost":1.25,
     "train_on_prompts":false,
     "hf_id":null,
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1771459200000.0,
+    "creation_date":1762992000000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -378,16 +357,16 @@
     ]
   },
   {
-    "id":"meta-llama\/llama-3-70b-instruct",
-    "name":"Llama 3 70B Instruct",
-    "provider_name":"Meta",
-    "cost":0.74,
+    "id":"deepseek\/deepseek-chat-v3.1",
+    "name":"DeepSeek V3.1",
+    "provider_name":"DeepSeek",
+    "cost":0.79,
     "train_on_prompts":false,
-    "hf_id":"meta-llama\/Meta-Llama-3-70B-Instruct",
-    "size":70553706496.0,
+    "hf_id":"deepseek-ai\/DeepSeek-V3.1",
+    "size":684531386000.0,
     "type":"open-source",
-    "license":"Llama3",
-    "creation_date":1713312000000.0,
+    "license":"Mit",
+    "creation_date":1755734400000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -399,16 +378,16 @@
     ]
   },
   {
-    "id":"meta-llama\/llama-3.1-70b-instruct",
-    "name":"Llama 3.1 70B Instruct",
-    "provider_name":"Meta",
-    "cost":0.4,
+    "id":"deepseek\/deepseek-v3.1-terminus",
+    "name":"DeepSeek V3.1 Terminus",
+    "provider_name":"DeepSeek",
+    "cost":0.95,
     "train_on_prompts":false,
-    "hf_id":"meta-llama\/Llama-3.1-70B-Instruct",
-    "size":70553706496.0,
+    "hf_id":"deepseek-ai\/DeepSeek-V3.1-Terminus",
+    "size":684531386000.0,
     "type":"open-source",
-    "license":"Llama3.1",
-    "creation_date":1721088000000.0,
+    "license":"Mit",
+    "creation_date":1758499200000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -420,16 +399,16 @@
     ]
   },
   {
-    "id":"meta-llama\/llama-3.3-70b-instruct",
-    "name":"Llama 3.3 70B Instruct",
-    "provider_name":"Meta",
-    "cost":0.32,
+    "id":"deepseek\/deepseek-v3.2-exp",
+    "name":"DeepSeek V3.2 Exp",
+    "provider_name":"DeepSeek",
+    "cost":0.41,
     "train_on_prompts":false,
-    "hf_id":"meta-llama\/Llama-3.3-70B-Instruct",
-    "size":null,
+    "hf_id":"deepseek-ai\/DeepSeek-V3.2-Exp",
+    "size":685396921376.0,
     "type":"open-source",
-    "license":"Llama3.3",
-    "creation_date":1732579200000.0,
+    "license":"Mit",
+    "creation_date":1759104000000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -441,16 +420,16 @@
     ]
   },
   {
-    "id":"meta-llama\/llama-4-maverick",
-    "name":"Llama 4 Maverick",
-    "provider_name":"Meta",
-    "cost":0.6,
+    "id":"deepseek\/deepseek-v4-flash-20260423",
+    "name":"DeepSeek V4 Flash",
+    "provider_name":"DeepSeek",
+    "cost":0.2,
     "train_on_prompts":false,
-    "hf_id":"meta-llama\/Llama-4-Maverick-17B-128E-Instruct",
-    "size":401583781376.0,
+    "hf_id":"deepseek-ai\/DeepSeek-V4-Flash",
+    "size":158069433298.0,
     "type":"open-source",
-    "license":"Other",
-    "creation_date":1743465600000.0,
+    "license":"Mit",
+    "creation_date":1776816000000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -462,16 +441,16 @@
     ]
   },
   {
-    "id":"microsoft\/phi-4",
-    "name":"Phi 4",
-    "provider_name":"Microsoft",
-    "cost":0.14,
+    "id":"google\/gemini-2.5-flash",
+    "name":"Gemini 2.5 Flash",
+    "provider_name":"Google",
+    "cost":2.5,
     "train_on_prompts":false,
-    "hf_id":"microsoft\/phi-4",
+    "hf_id":null,
     "size":null,
-    "type":"open-source",
-    "license":"Mit",
-    "creation_date":1733875200000.0,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1750118400000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -483,16 +462,16 @@
     ]
   },
   {
-    "id":"minimax\/minimax-m2.5",
-    "name":"MiniMax M2.5",
-    "provider_name":"MiniMax",
-    "cost":1.2,
+    "id":"google\/gemini-2.5-flash-lite",
+    "name":"Gemini 2.5 Flash Lite",
+    "provider_name":"Google",
+    "cost":0.4,
     "train_on_prompts":false,
-    "hf_id":"MiniMaxAI\/MiniMax-M2.5",
-    "size":228703644928.0,
-    "type":"open-source",
-    "license":"Other",
-    "creation_date":1770854400000.0,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1753142400000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -504,16 +483,16 @@
     ]
   },
   {
-    "id":"mistralai\/mistral-medium-3.1",
-    "name":"Mistral Medium 3.1",
-    "provider_name":"Mistral",
-    "cost":2.0,
+    "id":"google\/gemini-2.5-pro",
+    "name":"Gemini 2.5 Pro",
+    "provider_name":"Google",
+    "cost":10.0,
     "train_on_prompts":false,
     "hf_id":null,
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1755043200000.0,
+    "creation_date":1750118400000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -525,16 +504,16 @@
     ]
   },
   {
-    "id":"mistralai\/mistral-nemo",
-    "name":"Mistral Nemo",
-    "provider_name":"Mistral",
-    "cost":0.04,
+    "id":"google\/gemini-3.1-flash-lite",
+    "name":"Gemini 3.1 Flash Lite",
+    "provider_name":"Google",
+    "cost":1.5,
     "train_on_prompts":false,
-    "hf_id":"mistralai\/Mistral-Nemo-Instruct-2407",
+    "hf_id":null,
     "size":null,
-    "type":"open-source",
-    "license":"Apache 2.0",
-    "creation_date":1721174400000.0,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1778112000000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -546,16 +525,16 @@
     ]
   },
   {
-    "id":"mistralai\/mistral-saba",
-    "name":"Saba",
-    "provider_name":"Mistral",
-    "cost":0.6,
+    "id":"google\/gemini-3.1-pro-preview",
+    "name":"Gemini 3.1 Pro Preview",
+    "provider_name":"Google",
+    "cost":12.0,
     "train_on_prompts":false,
     "hf_id":null,
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1739750400000.0,
+    "creation_date":1771459200000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -567,16 +546,16 @@
     ]
   },
   {
-    "id":"moonshotai\/kimi-k2",
-    "name":"Kimi K2 0711",
-    "provider_name":"MoonshotAI",
-    "cost":2.2,
+    "id":"google\/gemma-3-27b-it",
+    "name":"Gemma 3 27B",
+    "provider_name":"Google",
+    "cost":0.16,
     "train_on_prompts":false,
-    "hf_id":"moonshotai\/Kimi-K2-Instruct",
-    "size":1026470731056.0,
+    "hf_id":"google\/gemma-3-27b-it",
+    "size":27432406640.0,
     "type":"open-source",
-    "license":"Other",
-    "creation_date":1752192000000.0,
+    "license":"Gemma",
+    "creation_date":1740787200000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -588,16 +567,16 @@
     ]
   },
   {
-    "id":"openai\/gpt-3.5-turbo",
-    "name":"GPT-3.5 Turbo",
-    "provider_name":"OpenAI",
-    "cost":1.5,
+    "id":"google\/gemma-4-26b-a4b-it-20260403",
+    "name":"Gemma 4 26B A4B ",
+    "provider_name":"Google",
+    "cost":0.33,
     "train_on_prompts":false,
-    "hf_id":null,
-    "size":null,
-    "type":"closed-source",
-    "license":null,
-    "creation_date":1685232000000.0,
+    "hf_id":"google\/gemma-4-26B-A4B-it",
+    "size":26544131376.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1773187200000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -609,16 +588,16 @@
     ]
   },
   {
-    "id":"openai\/gpt-4.1",
-    "name":"GPT-4.1",
-    "provider_name":"OpenAI",
-    "cost":8.0,
+    "id":"google\/gemma-4-31b-it-20260402",
+    "name":"Gemma 4 31B",
+    "provider_name":"Google",
+    "cost":0.37,
     "train_on_prompts":false,
-    "hf_id":null,
-    "size":null,
-    "type":"closed-source",
-    "license":null,
-    "creation_date":1744588800000.0,
+    "hf_id":"google\/gemma-4-31B-it",
+    "size":32682372656.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1773187200000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -630,16 +609,16 @@
     ]
   },
   {
-    "id":"openai\/gpt-4o",
-    "name":"GPT-4o",
-    "provider_name":"OpenAI",
-    "cost":10.0,
+    "id":"ibm-granite\/granite-4.0-h-micro",
+    "name":"Granite 4.0 Micro",
+    "provider_name":"IBM",
+    "cost":0.11,
     "train_on_prompts":false,
-    "hf_id":null,
-    "size":null,
-    "type":"closed-source",
-    "license":null,
-    "creation_date":1715558400000.0,
+    "hf_id":"ibm-granite\/granite-4.0-h-micro",
+    "size":3191396096.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1757980800000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -651,16 +630,16 @@
     ]
   },
   {
-    "id":"openai\/gpt-5",
-    "name":"GPT-5",
-    "provider_name":"OpenAI",
-    "cost":10.0,
+    "id":"ibm-granite\/granite-4.1-8b-20260429",
+    "name":"Granite 4.1 8B",
+    "provider_name":"IBM",
+    "cost":0.1,
     "train_on_prompts":false,
-    "hf_id":null,
-    "size":null,
-    "type":"closed-source",
-    "license":null,
-    "creation_date":1754524800000.0,
+    "hf_id":"ibm-granite\/granite-4.1-8b",
+    "size":8791592960.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1775433600000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -672,16 +651,16 @@
     ]
   },
   {
-    "id":"openai\/gpt-5-mini",
-    "name":"GPT-5 Mini",
-    "provider_name":"OpenAI",
-    "cost":2.0,
+    "id":"liquid\/lfm-2-24b-a2b-20260224",
+    "name":"LFM2-24B-A2B",
+    "provider_name":"LiquidAI",
+    "cost":0.12,
     "train_on_prompts":false,
-    "hf_id":null,
-    "size":null,
-    "type":"closed-source",
-    "license":null,
-    "creation_date":1754524800000.0,
+    "hf_id":"LiquidAI\/LFM2-24B-A2B",
+    "size":23843661440.0,
+    "type":"open-source",
+    "license":"Other",
+    "creation_date":1771891200000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -693,16 +672,16 @@
     ]
   },
   {
-    "id":"openai\/gpt-5-nano",
-    "name":"GPT-5 Nano",
-    "provider_name":"OpenAI",
-    "cost":0.4,
+    "id":"meta-llama\/llama-3-70b-instruct",
+    "name":"Llama 3 70B Instruct",
+    "provider_name":"Meta",
+    "cost":0.74,
     "train_on_prompts":false,
-    "hf_id":null,
-    "size":null,
-    "type":"closed-source",
-    "license":null,
-    "creation_date":1754524800000.0,
+    "hf_id":"meta-llama\/Meta-Llama-3-70B-Instruct",
+    "size":70553706496.0,
+    "type":"open-source",
+    "license":"Llama3",
+    "creation_date":1713312000000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -714,16 +693,184 @@
     ]
   },
   {
-    "id":"openai\/gpt-5.1",
-    "name":"GPT-5.1",
-    "provider_name":"OpenAI",
-    "cost":10.0,
+    "id":"meta-llama\/llama-3.1-70b-instruct",
+    "name":"Llama 3.1 70B Instruct",
+    "provider_name":"Meta",
+    "cost":0.4,
+    "train_on_prompts":false,
+    "hf_id":"meta-llama\/Llama-3.1-70B-Instruct",
+    "size":70553706496.0,
+    "type":"open-source",
+    "license":"Llama3.1",
+    "creation_date":1721088000000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"meta-llama\/llama-3.3-70b-instruct",
+    "name":"Llama 3.3 70B Instruct",
+    "provider_name":"Meta",
+    "cost":0.32,
+    "train_on_prompts":false,
+    "hf_id":"meta-llama\/Llama-3.3-70B-Instruct",
+    "size":70553706496.0,
+    "type":"open-source",
+    "license":"Llama3.3",
+    "creation_date":1732579200000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"meta-llama\/llama-4-maverick",
+    "name":"Llama 4 Maverick",
+    "provider_name":"Meta",
+    "cost":0.6,
+    "train_on_prompts":false,
+    "hf_id":"meta-llama\/Llama-4-Maverick-17B-128E-Instruct",
+    "size":401583781376.0,
+    "type":"open-source",
+    "license":"Other",
+    "creation_date":1743465600000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"microsoft\/phi-4",
+    "name":"Phi 4",
+    "provider_name":"Microsoft",
+    "cost":0.14,
+    "train_on_prompts":false,
+    "hf_id":"microsoft\/phi-4",
+    "size":14659507200.0,
+    "type":"open-source",
+    "license":"Mit",
+    "creation_date":1733875200000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"microsoft\/phi-4-mini-instruct",
+    "name":"Phi 4 Mini Instruct",
+    "provider_name":"Microsoft",
+    "cost":0.35,
+    "train_on_prompts":false,
+    "hf_id":"microsoft\/Phi-4-mini-instruct",
+    "size":3836021760.0,
+    "type":"open-source",
+    "license":"Mit",
+    "creation_date":1739923200000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"mistralai\/ministral-14b-2512",
+    "name":"Ministral 3 14B 2512",
+    "provider_name":"Mistral",
+    "cost":0.2,
+    "train_on_prompts":false,
+    "hf_id":"mistralai\/Ministral-3-14B-Instruct-2512",
+    "size":13945032240.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1761868800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"mistralai\/ministral-3b-2512",
+    "name":"Ministral 3 3B 2512",
+    "provider_name":"Mistral",
+    "cost":0.1,
+    "train_on_prompts":false,
+    "hf_id":"mistralai\/Ministral-3-3B-Instruct-2512",
+    "size":3849090048.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1761868800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"mistralai\/ministral-8b-2512",
+    "name":"Ministral 3 8B 2512",
+    "provider_name":"Mistral",
+    "cost":0.15,
+    "train_on_prompts":false,
+    "hf_id":"mistralai\/Ministral-3-8B-Instruct-2512",
+    "size":8918026716.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1761868800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"mistralai\/mistral-large-2512",
+    "name":"Mistral Large 3 2512",
+    "provider_name":"Mistral",
+    "cost":1.5,
     "train_on_prompts":false,
     "hf_id":null,
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1762992000000.0,
+    "creation_date":1764547200000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -735,16 +882,16 @@
     ]
   },
   {
-    "id":"openai\/gpt-5.2",
-    "name":"GPT-5.2",
-    "provider_name":"OpenAI",
-    "cost":14.0,
+    "id":"mistralai\/mistral-medium-3.1",
+    "name":"Mistral Medium 3.1",
+    "provider_name":"Mistral",
+    "cost":2.0,
     "train_on_prompts":false,
     "hf_id":null,
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1765324800000.0,
+    "creation_date":1755043200000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -756,16 +903,37 @@
     ]
   },
   {
-    "id":"openai\/gpt-5.4",
-    "name":"GPT-5.4",
-    "provider_name":"OpenAI",
-    "cost":15.0,
+    "id":"mistralai\/mistral-nemo",
+    "name":"Mistral Nemo",
+    "provider_name":"Mistral",
+    "cost":0.03,
+    "train_on_prompts":false,
+    "hf_id":"mistralai\/Mistral-Nemo-Instruct-2407",
+    "size":12247782400.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1721174400000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"mistralai\/mistral-saba",
+    "name":"Saba",
+    "provider_name":"Mistral",
+    "cost":0.6,
     "train_on_prompts":false,
     "hf_id":null,
     "size":null,
     "type":"closed-source",
     "license":null,
-    "creation_date":1772668800000.0,
+    "creation_date":1739750400000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -777,16 +945,16 @@
     ]
   },
   {
-    "id":"openai\/gpt-oss-120b",
-    "name":"gpt-oss-120b",
-    "provider_name":"OpenAI",
-    "cost":0.19,
+    "id":"mistralai\/mistral-small-2603",
+    "name":"Mistral Small 4",
+    "provider_name":"Mistral",
+    "cost":0.6,
     "train_on_prompts":false,
-    "hf_id":"openai\/gpt-oss-120b",
-    "size":120412337472.0,
+    "hf_id":"mistralai\/Mistral-Small-4-119B-2603",
+    "size":119401317952.0,
     "type":"open-source",
     "license":"Apache 2.0",
-    "creation_date":1754265600000.0,
+    "creation_date":1769126400000,
     "tasks":[
       "translation_from",
       "translation_to",
@@ -798,16 +966,940 @@
     ]
   },
   {
-    "id":"qwen\/qwen3-30b-a3b",
-    "name":"Qwen3 30B A3B",
-    "provider_name":"Qwen",
-    "cost":0.28,
+    "id":"mistralai\/mistral-small-3.2-24b-instruct",
+    "name":"Mistral Small 3.2 24B",
+    "provider_name":"Mistral",
+    "cost":0.2,
     "train_on_prompts":false,
-    "hf_id":"Qwen\/Qwen3-30B-A3B",
-    "size":null,
+    "hf_id":"mistralai\/Mistral-Small-3.2-24B-Instruct-2506",
+    "size":24011361280.0,
     "type":"open-source",
     "license":"Apache 2.0",
-    "creation_date":1745712000000.0,
+    "creation_date":1750291200000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"moonshotai\/kimi-k2",
+    "name":"Kimi K2 0711",
+    "provider_name":"MoonshotAI",
+    "cost":2.3,
+    "train_on_prompts":false,
+    "hf_id":"moonshotai\/Kimi-K2-Instruct",
+    "size":1026470731056.0,
+    "type":"open-source",
+    "license":"Other",
+    "creation_date":1752192000000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"moonshotai\/kimi-k2.6",
+    "name":"Kimi K2.6",
+    "provider_name":"MoonshotAI",
+    "cost":3.42,
+    "train_on_prompts":false,
+    "hf_id":"moonshotai\/Kimi-K2.6",
+    "size":1058589420528.0,
+    "type":"open-source",
+    "license":"Other",
+    "creation_date":1776124800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"nousresearch\/hermes-4-405b",
+    "name":"Hermes 4 405B",
+    "provider_name":"Nous",
+    "cost":3.0,
+    "train_on_prompts":false,
+    "hf_id":"NousResearch\/Hermes-4-405B",
+    "size":405853388800.0,
+    "type":"open-source",
+    "license":"Llama3",
+    "creation_date":1754438400000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"nousresearch\/hermes-4-70b",
+    "name":"Hermes 4 70B",
+    "provider_name":"Nous",
+    "cost":0.4,
+    "train_on_prompts":false,
+    "hf_id":"NousResearch\/Hermes-4-70B",
+    "size":70553706496.0,
+    "type":"open-source",
+    "license":"Llama3",
+    "creation_date":1755475200000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"nvidia\/llama-3.3-nemotron-super-49b-v1.5",
+    "name":"Llama 3.3 Nemotron Super 49B V1.5",
+    "provider_name":"NVIDIA",
+    "cost":0.4,
+    "train_on_prompts":false,
+    "hf_id":"nvidia\/Llama-3_3-Nemotron-Super-49B-v1_5",
+    "size":49867145216.0,
+    "type":"open-source",
+    "license":"Other",
+    "creation_date":1753401600000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"nvidia\/nemotron-3-nano-30b-a3b",
+    "name":"Nemotron 3 Nano 30B A3B",
+    "provider_name":"NVIDIA",
+    "cost":0.2,
+    "train_on_prompts":false,
+    "hf_id":"nvidia\/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16",
+    "size":31577937344.0,
+    "type":"open-source",
+    "license":"Other",
+    "creation_date":1764806400000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"nvidia\/nemotron-3-super-120b-a12b-20230311",
+    "name":"Nemotron 3 Super",
+    "provider_name":"NVIDIA",
+    "cost":0.45,
+    "train_on_prompts":false,
+    "hf_id":"nvidia\/NVIDIA-Nemotron-3-Super-120B-A12B-FP8",
+    "size":123611012096.0,
+    "type":"open-source",
+    "license":"Other",
+    "creation_date":1773100800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"nvidia\/nemotron-3-ultra-550b-a55b-20260604",
+    "name":"Nemotron 3 Ultra",
+    "provider_name":"NVIDIA",
+    "cost":2.5,
+    "train_on_prompts":false,
+    "hf_id":"nvidia\/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
+    "size":560524578816.0,
+    "type":"open-source",
+    "license":"Other",
+    "creation_date":1780444800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"nvidia\/nemotron-nano-9b-v2",
+    "name":"Nemotron Nano 9B V2",
+    "provider_name":"NVIDIA",
+    "cost":0.16,
+    "train_on_prompts":false,
+    "hf_id":"nvidia\/NVIDIA-Nemotron-Nano-9B-v2",
+    "size":8888227328.0,
+    "type":"open-source",
+    "license":"Other",
+    "creation_date":1754956800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-3.5-turbo",
+    "name":"GPT-3.5 Turbo",
+    "provider_name":"OpenAI",
+    "cost":1.5,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1685232000000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-4.1",
+    "name":"GPT-4.1",
+    "provider_name":"OpenAI",
+    "cost":8.0,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1744588800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-4o",
+    "name":"GPT-4o",
+    "provider_name":"OpenAI",
+    "cost":10.0,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1715558400000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-5",
+    "name":"GPT-5",
+    "provider_name":"OpenAI",
+    "cost":10.0,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1754524800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-5-mini",
+    "name":"GPT-5 Mini",
+    "provider_name":"OpenAI",
+    "cost":2.0,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1754524800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-5-nano",
+    "name":"GPT-5 Nano",
+    "provider_name":"OpenAI",
+    "cost":0.4,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1754524800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-5.1",
+    "name":"GPT-5.1",
+    "provider_name":"OpenAI",
+    "cost":10.0,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1762992000000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-5.2",
+    "name":"GPT-5.2",
+    "provider_name":"OpenAI",
+    "cost":14.0,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1765324800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-5.3-chat-20260303",
+    "name":"GPT-5.3 Chat",
+    "provider_name":"OpenAI",
+    "cost":14.0,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1772496000000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-5.4",
+    "name":"GPT-5.4",
+    "provider_name":"OpenAI",
+    "cost":15.0,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1772668800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-5.5",
+    "name":"GPT-5.5",
+    "provider_name":"OpenAI",
+    "cost":30.0,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1776988800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-oss-120b",
+    "name":"gpt-oss-120b",
+    "provider_name":"OpenAI",
+    "cost":0.18,
+    "train_on_prompts":false,
+    "hf_id":"openai\/gpt-oss-120b",
+    "size":120412337472.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1754265600000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"openai\/gpt-oss-20b",
+    "name":"gpt-oss-20b",
+    "provider_name":"OpenAI",
+    "cost":0.14,
+    "train_on_prompts":false,
+    "hf_id":"openai\/gpt-oss-20b",
+    "size":21511953984.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1754265600000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"prime-intellect\/intellect-3-20251126",
+    "name":"INTELLECT-3",
+    "provider_name":"Prime Intellect",
+    "cost":1.1,
+    "train_on_prompts":false,
+    "hf_id":"PrimeIntellect\/INTELLECT-3-FP8",
+    "size":106893249280.0,
+    "type":"open-source",
+    "license":"Mit",
+    "creation_date":1764115200000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"qwen\/qwen3-30b-a3b",
+    "name":"Qwen3 30B A3B",
+    "provider_name":"Qwen",
+    "cost":0.45,
+    "train_on_prompts":false,
+    "hf_id":"Qwen\/Qwen3-30B-A3B",
+    "size":30532122624.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1745712000000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"qwen\/qwen3-30b-a3b-instruct-2507",
+    "name":"Qwen3 30B A3B Instruct 2507",
+    "provider_name":"Qwen",
+    "cost":0.19,
+    "train_on_prompts":false,
+    "hf_id":"Qwen\/Qwen3-30B-A3B-Instruct-2507",
+    "size":30532122624.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1753660800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"qwen\/qwen3-max",
+    "name":"Qwen3 Max",
+    "provider_name":"Qwen",
+    "cost":3.9,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1758585600000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"qwen\/qwen3-next-80b-a3b-instruct-2509",
+    "name":"Qwen3 Next 80B A3B Instruct",
+    "provider_name":"Qwen",
+    "cost":1.1,
+    "train_on_prompts":false,
+    "hf_id":"Qwen\/Qwen3-Next-80B-A3B-Instruct",
+    "size":81324862720.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1757376000000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"qwen\/qwen3.5-122b-a10b-20260224",
+    "name":"Qwen3.5-122B-A10B",
+    "provider_name":"Qwen",
+    "cost":2.08,
+    "train_on_prompts":false,
+    "hf_id":"Qwen\/Qwen3.5-122B-A10B",
+    "size":125086497008.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1771891200000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"qwen\/qwen3.5-397b-a17b-20260216",
+    "name":"Qwen3.5 397B A17B",
+    "provider_name":"Qwen",
+    "cost":2.34,
+    "train_on_prompts":false,
+    "hf_id":"Qwen\/Qwen3.5-397B-A17B",
+    "size":403397928944.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1771200000000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"qwen\/qwen3.5-9b-20260310",
+    "name":"Qwen3.5-9B",
+    "provider_name":"Qwen",
+    "cost":0.15,
+    "train_on_prompts":false,
+    "hf_id":"Qwen\/Qwen3.5-9B",
+    "size":9653104368.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1772150400000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"qwen\/qwen3.6-27b-20260422",
+    "name":"Qwen3.6 27B",
+    "provider_name":"Qwen",
+    "cost":3.2,
+    "train_on_prompts":false,
+    "hf_id":"Qwen\/Qwen3.6-27B",
+    "size":27781427952.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1776729600000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"qwen\/qwen3.6-35b-a3b-20260415",
+    "name":"Qwen3.6 35B A3B",
+    "provider_name":"Qwen",
+    "cost":1.0,
+    "train_on_prompts":false,
+    "hf_id":"Qwen\/Qwen3.6-35B-A3B",
+    "size":35951822704.0,
+    "type":"open-source",
+    "license":"Apache 2.0",
+    "creation_date":1776211200000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"qwen\/qwen3.6-flash",
+    "name":"Qwen3.6 Flash",
+    "provider_name":"Qwen",
+    "cost":1.12,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1777248000000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"qwen\/qwen3.6-plus-04-02",
+    "name":"Qwen3.6 Plus",
+    "provider_name":"Qwen",
+    "cost":1.95,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1775088000000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"rekaai\/reka-edge-2603",
+    "name":"Reka Edge",
+    "provider_name":"Reka Edge",
+    "cost":0.1,
+    "train_on_prompts":false,
+    "hf_id":"RekaAI\/reka-edge-2603",
+    "size":7128509568.0,
+    "type":"open-source",
+    "license":"Other",
+    "creation_date":1773187200000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"stepfun\/step-3.7-flash-20260528",
+    "name":"Step 3.7 Flash",
+    "provider_name":"StepFun",
+    "cost":1.15,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1779926400000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"tencent\/hunyuan-a13b-instruct",
+    "name":"Hunyuan A13B Instruct",
+    "provider_name":"Tencent",
+    "cost":0.57,
+    "train_on_prompts":false,
+    "hf_id":"tencent\/Hunyuan-A13B-Instruct",
+    "size":80393183232.0,
+    "type":"open-source",
+    "license":"Other",
+    "creation_date":1750809600000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"upstage\/solar-pro-3",
+    "name":"Solar Pro 3",
+    "provider_name":"Upstage",
+    "cost":0.6,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1769472000000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"writer\/palmyra-x5-20250428",
+    "name":"Palmyra X5",
+    "provider_name":"Writer",
+    "cost":6.0,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1768953600000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"x-ai\/grok-4.20",
+    "name":"Grok 4.20",
+    "provider_name":"xAI",
+    "cost":2.5,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1774915200000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"z-ai\/glm-4-32b-0414",
+    "name":"GLM 4 32B ",
+    "provider_name":"Z.ai",
+    "cost":0.1,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1753315200000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"z-ai\/glm-4.5-air",
+    "name":"GLM 4.5 Air",
+    "provider_name":"Z.ai",
+    "cost":0.85,
+    "train_on_prompts":false,
+    "hf_id":"zai-org\/GLM-4.5-Air",
+    "size":110468824832.0,
+    "type":"open-source",
+    "license":"Mit",
+    "creation_date":1752969600000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"z-ai\/glm-4.7-flash-20260119",
+    "name":"GLM 4.7 Flash",
+    "provider_name":"Z.ai",
+    "cost":0.4,
+    "train_on_prompts":false,
+    "hf_id":"zai-org\/GLM-4.7-Flash",
+    "size":31221488576.0,
+    "type":"open-source",
+    "license":"Mit",
+    "creation_date":1768780800000,
+    "tasks":[
+      "translation_from",
+      "translation_to",
+      "classification",
+      "mmlu",
+      "arc",
+      "truthfulqa",
+      "mgsm"
+    ]
+  },
+  {
+    "id":"z-ai\/glm-5-turbo-20260315",
+    "name":"GLM 5 Turbo",
+    "provider_name":"Z.ai",
+    "cost":4.0,
+    "train_on_prompts":false,
+    "hf_id":null,
+    "size":null,
+    "type":"closed-source",
+    "license":null,
+    "creation_date":1773532800000,
     "tasks":[
       "translation_from",
       "translation_to",

results/results.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:75c0a0374e1e03392d76db5baf38e282b6af6dec080fab2b9cf1af7e68c931c8
-size 9437932
+oid sha256:03f2b50b3b400f8bf42254f579f68faaa07c20cfe0978a604fdd06b7581450ac
+size 16911217