Update data.json

data.json

@@ -5,15 +5,23 @@
         "Sentiment": 4.230769230769231,
         "Language understanding": 4.0,
         "Phraseology": 3.86,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.646067415730337
     },
     {
-        "Model": "alpindale/WizardLM-2-8x22B",
+        "Model": "mistralai/Mistral-Large-Instruct-2411",
+        "Params": "123B",
+        "Sentiment": 4.326923076923077,
+        "Language understanding": 3.975,
+        "Phraseology": 3.99,
+        "Tricky questions": 3.7247191011235956
+    },
+    {
+        "Model": "alpindale/WizardLM-2-8x22B (API)",
         "Params": "141B",
         "Sentiment": 3.7051282051282053,
         "Language understanding": 3.815,
         "Phraseology": 4.22,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.056179775280899
     },
     {
         "Model": "meta-llama/Meta-Llama-3.1-70B-Instruct",
@@ -21,7 +29,7 @@
         "Sentiment": 4.326923076923077,
         "Language understanding": 3.91,
         "Phraseology": 3.25,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.0112359550561796
     },
     {
         "Model": "meta-llama/Meta-Llama-3-70B-Instruct",
@@ -29,7 +37,7 @@
         "Sentiment": 4.134615384615385,
         "Language understanding": 3.82,
         "Phraseology": 3.465,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.707865168539326
     },
     {
         "Model": "speakleash/Bielik-11B-v2.3-Instruct",
@@ -37,15 +45,15 @@
         "Sentiment": 3.9743589743589745,
         "Language understanding": 3.785,
         "Phraseology": 3.55,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.2191011235955056
     },
     {
-        "Model": "mistralai/Mixtral-8x22B-Instruct-v0.1",
+        "Model": "mistralai/Mixtral-8x22B-Instruct-v0.1 (API)",
         "Params": "141B",
         "Sentiment": 3.782051282051282,
         "Language understanding": 3.675,
         "Phraseology": 3.55,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.235955056179775
     },
     {
         "Model": "speakleash/Bielik-11B-v2.1-Instruct",
@@ -53,7 +61,7 @@
         "Sentiment": 3.9551282051282053,
         "Language understanding": 3.915,
         "Phraseology": 3.105,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.4719101123595504
     },
     {
         "Model": "Qwen/Qwen2-72B-Instruct",
@@ -61,7 +69,7 @@
         "Sentiment": 3.7628205128205128,
         "Language understanding": 3.89,
         "Phraseology": 3.28,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.6797752808988764
     },
     {
         "Model": "speakleash/Bielik-11B-v2.0-Instruct",
@@ -69,7 +77,7 @@
         "Sentiment": 3.9743589743589745,
         "Language understanding": 3.745,
         "Phraseology": 3.125,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.196629213483146
     },
     {
         "Model": "speakleash/Bielik-11B-v2.2-Instruct",
@@ -77,7 +85,7 @@
         "Sentiment": 3.717948717948718,
         "Language understanding": 3.73,
         "Phraseology": 3.25,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.1235955056179776
     },
     {
         "Model": "Qwen/Qwen1.5-72B-Chat",
@@ -85,7 +93,7 @@
         "Sentiment": 3.4743589743589745,
         "Language understanding": 3.515,
         "Phraseology": 2.975,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.668539325842697
     },
     {
         "Model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
@@ -93,7 +101,7 @@
         "Sentiment": 3.9743589743589745,
         "Language understanding": 3.38,
         "Phraseology": 2.58,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.1123595505617976
     },
     {
         "Model": "THUDM/glm-4-9b-chat",
@@ -101,7 +109,7 @@
         "Sentiment": 3.58974358974359,
         "Language understanding": 3.455,
         "Phraseology": 2.78,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.9831460674157304
     },
     {
         "Model": "mistralai/Mistral-Nemo-Instruct-2407",
@@ -109,7 +117,7 @@
         "Sentiment": 3.641025641025641,
         "Language understanding": 3.29,
         "Phraseology": 2.74,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.0898876404494384
     },
     {
         "Model": "meta-llama/Meta-Llama-3-8B-Instruct",
@@ -117,7 +125,7 @@
         "Sentiment": 3.3333333333333335,
         "Language understanding": 3.15,
         "Phraseology": 3.035,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.4775280898876404
     },
     {
         "Model": "upstage/SOLAR-10.7B-Instruct-v1.0",
@@ -125,7 +133,7 @@
         "Sentiment": 2.967948717948718,
         "Language understanding": 3.18,
         "Phraseology": 3.255,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.1235955056179776
     },
     {
         "Model": "speakleash/Bielik-7B-Instruct-v0.1",
@@ -133,7 +141,7 @@
         "Sentiment": 3.58974358974359,
         "Language understanding": 3.475,
         "Phraseology": 2.315,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.157303370786517
     },
     {
         "Model": "openchat/openchat-3.5-0106-gemma",
@@ -141,7 +149,7 @@
         "Sentiment": 3.730769230769231,
         "Language understanding": 3.08,
         "Phraseology": 2.445,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.6797752808988764
     },
     {
         "Model": "mistralai/Mixtral-8x7B-Instruct-v0.1",
@@ -149,7 +157,7 @@
         "Sentiment": 3.0576923076923075,
         "Language understanding": 3.175,
         "Phraseology": 2.885,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.797752808988764
     },
     {
         "Model": "mistralai/Mistral-7B-Instruct-v0.3",
@@ -157,7 +165,7 @@
         "Sentiment": 3.326923076923077,
         "Language understanding": 3.06,
         "Phraseology": 2.68,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.9887640449438202
     },
     {
         "Model": "berkeley-nest/Starling-LM-7B-alpha",
@@ -165,7 +173,7 @@
         "Sentiment": 3.0576923076923075,
         "Language understanding": 2.925,
         "Phraseology": 2.855,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.6797752808988764
     },
     {
         "Model": "openchat/openchat-3.5-0106",
@@ -173,7 +181,7 @@
         "Sentiment": 3.16025641025641,
         "Language understanding": 2.835,
         "Phraseology": 2.555,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.9606741573033708
     },
     {
         "Model": "internlm/internlm2-chat-20b",
@@ -181,7 +189,7 @@
         "Sentiment": 3.301282051282051,
         "Language understanding": 2.785,
         "Phraseology": 2.385,
-        "Tricky questions": 3.9
+        "Tricky questions": 0.12359550561797752
     },
     {
         "Model": "01-ai/Yi-1.5-34B-Chat",
@@ -189,7 +197,7 @@
         "Sentiment": 3.076923076923077,
         "Language understanding": 2.87,
         "Phraseology": 2.38,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.0
     },
     {
         "Model": "Voicelab/trurl-2-13b-academic",
@@ -197,7 +205,7 @@
         "Sentiment": 3.301282051282051,
         "Language understanding": 2.755,
         "Phraseology": 2.165,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.0168539325842696
     },
     {
         "Model": "google/gemma-2-2b-it",
@@ -205,7 +213,7 @@
         "Sentiment": 3.3974358974359,
         "Language understanding": 2.9,
         "Phraseology": 2.095,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.2134831460674156
     },
     {
         "Model": "Qwen/Qwen2.5-3B-Instruct",
@@ -213,7 +221,7 @@
         "Sentiment": 2.948717948717949,
         "Language understanding": 2.455,
         "Phraseology": 2.8,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.8089887640449438
     },
     {
         "Model": "NousResearch/Hermes-3-Llama-3.2-3B",
@@ -221,7 +229,7 @@
         "Sentiment": 2.6153846153846154,
         "Language understanding": 2.705,
         "Phraseology": 2.765,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.1404494382022472
     },
     {
         "Model": "ibm-granite/granite-3.1-2b-instruct",
@@ -229,7 +237,7 @@
         "Sentiment": 3.076923076923077,
         "Language understanding": 2.235,
         "Phraseology": 1.88,
-        "Tricky questions": 3.9
+        "Tricky questions": 0.5898876404494382
     },
     {
         "Model": "meta-llama/Llama-3.2-1B-Instruct",
@@ -237,7 +245,7 @@
         "Sentiment": 3.076923076923077,
         "Language understanding": 1.735,
         "Phraseology": 2.34,
-        "Tricky questions": 3.9
+        "Tricky questions": 0.5224719101123596
     },
     {
         "Model": "microsoft/Phi-3.5-mini-instruct",
@@ -245,7 +253,7 @@
         "Sentiment": 2.435897435897436,
         "Language understanding": 2.135,
         "Phraseology": 2.425,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.0449438202247192
     },
     {
         "Model": "meta-llama/Llama-3.2-3B-Instruct",
@@ -253,7 +261,7 @@
         "Sentiment": 2.7564102564102564,
         "Language understanding": 2.295,
         "Phraseology": 1.72,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.2191011235955056
     },
     {
         "Model": "h2oai/h2o-danube2-1.8b-chat",
@@ -261,7 +269,7 @@
         "Sentiment": 2.371794871794872,
         "Language understanding": 1.595,
         "Phraseology": 2.47,
-        "Tricky questions": 3.9
+        "Tricky questions": 0.12921348314606743
     },
     {
         "Model": "Qwen/Qwen2.5-1.5B-Instruct",
@@ -269,7 +277,7 @@
         "Sentiment": 2.7948717948717947,
         "Language understanding": 1.35,
         "Phraseology": 2.225,
-        "Tricky questions": 3.9
+        "Tricky questions": 0.6629213483146067
     },
     {
         "Model": "utter-project/EuroLLM-1.7B-Instruct",
@@ -277,7 +285,7 @@
         "Sentiment": 2.243589743589744,
         "Language understanding": 1.79,
         "Phraseology": 2.26,
-        "Tricky questions": 3.9
+        "Tricky questions": 0.7584269662921348
     },
     {
         "Model": "LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct",
@@ -285,7 +293,7 @@
         "Sentiment": 1.9423076923076923,
         "Language understanding": 2.1155778894472363,
         "Phraseology": 2.130653266331658,
-        "Tricky questions": 3.9
+        "Tricky questions": 0.4887640449438202
     },
     {
         "Model": "HuggingFaceTB/SmolLM2-1.7B-Instruct",
@@ -293,7 +301,7 @@
         "Sentiment": 2.275641025641025,
         "Language understanding": 1.1,
         "Phraseology": 2.355,
-        "Tricky questions": 3.9
+        "Tricky questions": 0.25280898876404495
     },
     {
         "Model": "Qwen/Qwen2.5-0.5B-Instruct",
@@ -301,7 +309,7 @@
         "Sentiment": 1.955128205128205,
         "Language understanding": 0.835,
         "Phraseology": 2.595,
-        "Tricky questions": 3.9
+        "Tricky questions": 0.21910112359550563
     },
     {
         "Model": "CYFRAGOVPL/Llama-PLLuM-70B-chat",
@@ -309,7 +317,7 @@
         "Sentiment": 3.94,
         "Language understanding": 3.61,
         "Phraseology": 3.35,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.2134831460674156
     },
     {
         "Model": "CYFRAGOVPL/PLLuM-8x7B-nc-instruct",
@@ -317,7 +325,7 @@
         "Sentiment": 3.88,
         "Language understanding": 3.59,
         "Phraseology": 3.22,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.7640449438202248
     },
     {
         "Model": "CYFRAGOVPL/Llama-PLLuM-70B-instruct",
@@ -325,7 +333,7 @@
         "Sentiment": 3.78,
         "Language understanding": 3.63,
         "Phraseology": 3.26,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.634831460674157
     },
     {
         "Model": "CYFRAGOVPL/PLLuM-8x7B-instruct",
@@ -333,7 +341,7 @@
         "Sentiment": 3.59,
         "Language understanding": 3.47,
         "Phraseology": 3.46,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.5056179775280898
     },
     {
         "Model": "CYFRAGOVPL/PLLuM-12B-instruct",
@@ -341,7 +349,7 @@
         "Sentiment": 3.71,
         "Language understanding": 3.17,
         "Phraseology": 3.59,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.904494382022472
     },
     {
         "Model": "CYFRAGOVPL/PLLuM-8x7B-nc-chat",
@@ -349,7 +357,7 @@
         "Sentiment": 3.76,
         "Language understanding": 3.48,
         "Phraseology": 3.08,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.797752808988764
     },
     {
         "Model": "CYFRAGOVPL/PLLuM-8x7B-chat",
@@ -357,7 +365,7 @@
         "Sentiment": 3.44,
         "Language understanding": 3.45,
         "Phraseology": 3.35,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.7808988764044944
     },
     {
         "Model": "CYFRAGOVPL/PLLuM-12B-chat",
@@ -365,7 +373,7 @@
         "Sentiment": 3.32,
         "Language understanding": 3.21,
         "Phraseology": 3.43,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.5898876404494384
     },
     {
         "Model": "CYFRAGOVPL/PLLuM-12B-nc-instruct",
@@ -373,7 +381,7 @@
         "Sentiment": 3.24,
         "Language understanding": 3.31,
         "Phraseology": 3.32,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.9831460674157304
     },
     {
         "Model": "CYFRAGOVPL/Llama-PLLuM-8B-instruct",
@@ -381,7 +389,7 @@
         "Sentiment": 3.24,
         "Language understanding": 2.90,
         "Phraseology": 3.46,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.6629213483146068
     },
     {
         "Model": "CYFRAGOVPL/Llama-PLLuM-8B-chat",
@@ -389,7 +397,7 @@
         "Sentiment": 3.13,
         "Language understanding": 2.93,
         "Phraseology": 3.36,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.252808988764045
     },
     {
         "Model": "CYFRAGOVPL/PLLuM-12B-nc-chat",
@@ -397,7 +405,7 @@
         "Sentiment": 3.22,
         "Language understanding": 3.23,
         "Phraseology": 3.54,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.6235955056179776
     },
     {
         "Model": "Qwen/Qwen2.5-72B-Instruct",
@@ -405,7 +413,7 @@
         "Sentiment": 4.076923076923077,
         "Language understanding": 3.97,
         "Phraseology": 3.93,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.808988764044944
     },
     {
         "Model": "Qwen/Qwen2.5-32B-Instruct",
@@ -413,7 +421,7 @@
         "Sentiment": 3.8141025641025643,
         "Language understanding": 3.565,
         "Phraseology": 4.035,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.5898876404494384
     },
     {
         "Model": "mistralai/Mistral-Small-24B-Instruct-2501",
@@ -421,7 +429,7 @@
         "Sentiment": 3.91025641025641,
         "Language understanding": 3.6,
         "Phraseology": 3.875,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.449438202247191
     },
     {
         "Model": "meta-llama/Llama-3.3-70B-Instruct",
@@ -429,7 +437,7 @@
         "Sentiment": 4.294871794871795,
         "Language understanding": 3.865,
         "Phraseology": 3.04,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.3764044943820224
     },
     {
         "Model": "Qwen/Qwen2.5-14B-Instruct",
@@ -437,7 +445,7 @@
         "Sentiment": 3.91025641025641,
         "Language understanding": 3.565,
         "Phraseology": 3.37,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.337078651685393
     },
     {
         "Model": "microsoft/phi-4",
@@ -445,7 +453,7 @@
         "Sentiment": 3.717948717948718,
         "Language understanding": 3.54,
         "Phraseology": 3.235,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.7247191011235956
     },
     {
         "Model": "Qwen/Qwen2.5-7B-Instruct",
@@ -453,7 +461,7 @@
         "Sentiment": 3.5576923076923075,
         "Language understanding": 3.025,
         "Phraseology": 3.095,
-        "Tricky questions": 3.9
+        "Tricky questions": 2.5842696629213484
     },
     {
         "Model": "microsoft/Phi-4-mini-instruct",
@@ -461,7 +469,7 @@
         "Sentiment": 2.6923076923076925,
         "Language understanding": 2.43,
         "Phraseology": 2.245,
-        "Tricky questions": 3.9
+        "Tricky questions": 1.303370786516854
     },
     {
         "Model": "gemini-2.0-flash-001",
@@ -469,7 +477,7 @@
         "Sentiment": 4.519230769230769,
         "Language understanding": 4.32,
         "Phraseology": 4.34,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.9887640449438204
     },
     {
         "Model": "gemini-2.0-flash-lite-001",
@@ -477,7 +485,7 @@
         "Sentiment": 4.230769230769231,
         "Language understanding": 4.055,
         "Phraseology": 4.235,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.853932584269663
     },
     {
         "Model": "deepseek-ai/DeepSeek-V3 (API)",
@@ -485,7 +493,23 @@
         "Sentiment": 4.358974358974359,
         "Language understanding": 4.22,
         "Phraseology": 3.525,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.9887640449438204
+    },
+    {
+        "Model": "deepseek-ai/DeepSeek-R1 (API)",
+        "Params": "685B",
+        "Sentiment": 4.487179487179487,
+        "Language understanding": 4.345,
+        "Phraseology": 3.6,
+        "Tricky questions": 4.117977528089888
+    },
+    {
+        "Model": "deepseek-ai/DeepSeek-V3-0324 (API)",
+        "Params": "685B",
+        "Sentiment": 4.358974358974359,
+        "Language understanding": 4.195,
+        "Phraseology": 3.54,
+        "Tricky questions": 4.022471910112359
     },
     {
         "Model": "google/gemma-3-27b-it (API)",
@@ -493,6 +517,22 @@
         "Sentiment": 3.878205128205128,
         "Language understanding": 3.785,
         "Phraseology": 4.025,
-        "Tricky questions": 3.9
+        "Tricky questions": 3.533707865168539
+    },
+    {
+        "Model": "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8 (API)",
+        "Params": "402B",
+        "Sentiment": 4.391025641025641,
+        "Language understanding": 4.11,
+        "Phraseology": 3.475,
+        "Tricky questions": 3.758426966292135
+    },
+    {
+        "Model": "meta-llama/Llama-4-Scout-17B-16E-Instruct (API)",
+        "Params": "109B",
+        "Sentiment": 4.102564102564102,
+        "Language understanding": 3.805,
+        "Phraseology": 3.9,
+        "Tricky questions": 3.191011235955056
     }
-]
+]