update news and leaderboard data

app.py

@@ -914,6 +914,10 @@ def build_app():
                                 label="Comparative Analysis (Group Bar Chart)",
                                 elem_classes="visualization-container"
                             )
+                            gr.Markdown(
+                                    "*💡 Click a legend entry to isolate that model. Double-click additional ones to add them for comparison.*",
+                                    elem_classes="radar-tip"
+                                )
                             
 
                 # Hidden placeholder for group bar visualization (to maintain code references)
@@ -1143,6 +1147,10 @@ def build_app():
                                 label="Comparative Analysis (Group Bar Chart)",
                                 elem_classes="visualization-container"
                             )
+                            gr.Markdown(
+                                    "*💡 Click a legend entry to isolate that model. Double-click additional ones to add them for comparison.*",
+                                    elem_classes="radar-tip"
+                                )
 
                 # Game selection section
                 with gr.Row():

assets/game_video_link.json

@@ -1,7 +1,8 @@
 {
     "sokoban": "https://www.youtube.com/watch?v=59enV32MBUE",
-    "super_mario": "https://www.youtube.com/watch?v=nixMIJZYAgg",
+    "super_mario_bros": "https://www.youtube.com/watch?v=nixMIJZYAgg",
     "2048": "https://www.youtube.com/watch?v=3aYDCSa3AWI",
     "candy": "https://www.youtube.com/watch?v=b-Uyz3W4yIg",
-    "ace_attorney": "https://www.youtube.com/watch?v=q8PMW870yp8"
+    "ace_attorney": "https://www.youtube.com/watch?v=q8PMW870yp8",
+    "tetris": "https://www.youtube.com/watch?v=m6i9L6-pgu4"
 }

assets/model_color.json

@@ -1,6 +1,5 @@
 {
     "claude-3-7-sonnet-20250219": "#4A90E2",
-    "claude-3-7-sonnet-20250219 (thinking)": "#2E5C8A",
     "claude-3-5-haiku-20241022": "#7FB5E6",
     "claude-3-5-sonnet-20241022": "#1A4C7C",
     "claude-opus-4-20250514": "#3A80D2",
@@ -9,9 +8,8 @@
     "gemini-2.0-flash-thinking-exp-1219": "#C2185B",
     "gemini-2.5-pro-exp-03-25": "#FF80AB",
     "gemini-2.5-flash-preview-04-17": "#F06292",
-    "gemini-2.5-flash-preview-04-17 (thinking)": "#E91E63",
     "gemini-2.5-flash-preview-05-20": "#F8BBD9",
-    "gemini-2.5-pro-preview-05-06 (thinking)": "#AD1457",
+    "gemini-2.5-pro-preview-05-06": "#AD1457",
     "gemini-2.5-pro-preview-06-05": "#EC407A",
     "gpt-4o-2024-11-20": "#00BFA5",
     "gpt-4.5-preview-2025-02-27": "#00796B",
@@ -23,7 +21,6 @@
     "o4-mini-2025-04-16": "#00ACC1",
     "grok-3-beta": "#FF7043",
     "grok-3-mini-beta": "#FF8A65",
-    "grok-3-mini-beta (thinking)": "#F57C00",
     "deepseek-v3": "#FFC107",
     "deepseek-r1-0120": "#FFA000",
     "deepseek-r1-0528": "#FFB300",
@@ -31,7 +28,6 @@
     "qwen3-235B-A22B-fp8": "#6A1B9A",
     "random (x30)": "#9E9E9E",
     "gamingagent + claude-3-7-sonnet-20250219": "#4A90E2",
-    "gamingagent + claude-3-7-sonnet-20250219 (thinking)": "#2E5C8A",
     "gamingagent + claude-3-5-haiku-20241022": "#7FB5E6",
     "gamingagent + claude-3-5-sonnet-20241022": "#1A4C7C",
     "gamingagent + claude-opus-4-20250514": "#3A80D2",
@@ -40,9 +36,8 @@
     "gamingagent + gemini-2.0-flash-thinking-exp-1219": "#C2185B",
     "gamingagent + gemini-2.5-pro-exp-03-25": "#FF80AB",
     "gamingagent + gemini-2.5-flash-preview-04-17": "#F06292",
-    "gamingagent + gemini-2.5-flash-preview-04-17 (thinking)": "#E91E63",
     "gamingagent + gemini-2.5-flash-preview-05-20": "#F8BBD9",
-    "gamingagent + gemini-2.5-pro-preview-05-06 (thinking)": "#AD1457",
+    "gamingagent + gemini-2.5-pro-preview-05-06": "#AD1457",
     "gamingagent + gemini-2.5-pro-preview-06-05": "#EC407A",
     "gamingagent + gpt-4o-2024-11-20": "#00BFA5",
     "gamingagent + gpt-4.5-preview-2025-02-27": "#00796B",
@@ -54,7 +49,6 @@
     "gamingagent + o4-mini-2025-04-16": "#00ACC1",
     "gamingagent + grok-3-beta": "#FF7043",
     "gamingagent + grok-3-mini-beta": "#FF8A65",
-    "gamingagent + grok-3-mini-beta (thinking)": "#F57C00",
     "gamingagent + deepseek-v3": "#FFC107",
     "gamingagent + deepseek-r1-0120": "#FFA000",
     "gamingagent + deepseek-r1-0528": "#FFB300",

assets/news.json

@@ -1,10 +1,22 @@
 {
     "news": [
+        {
+            "date": "2025-06-07",
+            "video_link": "https://www.youtube.com/watch?v=4CV-fB5EVJs",
+            "twitter_text": "New Benchmark Results: Claude-sonnet-4 & Claude-opus-4 and Join Us in the Discord!",
+            "twitter_link": "https://x.com/haoailab/status/1931438794756313530"
+        },
+        {
+            "date": "2025-06-03",
+            "video_link": "https://www.youtube.com/watch?v=m6i9L6-pgu4",
+            "twitter_text": "New Benchmark Results: How do top open-source models like Deepseek r1 & Qwen 3 perform on games?",
+            "twitter_link": "https://x.com/haoailab/status/1929997363407708646"
+        },
         {
             "date": "2025-04-28",
             "video_link": "https://www.youtube.com/watch?v=OEQRhBKYxIE",
             "twitter_text": "Grok-3-mini-beta Joins the Battle: Outperforms Gemini 2.5 Flash, Challenges O3-mini Across Games — Full Grok-3-beta Power Yet to Come. 🚀",
-            "twitter_link": "https://x.com/haoailab"
+            "twitter_link": "https://x.com/haoailab/status/1917309598861779021"
         },
         {
             "date": "2025-04-24",

data_visualization.py

@@ -314,6 +314,7 @@ def create_group_bar_chart(df, top_n=10):
         uniformtext=dict(mode='hide', minsize=8),  # Hide text that doesn't fit
         legend=dict(
             font=dict(size=12),
+            title="Choose your model 💡 (click / double-click)",
             itemsizing='trace',
             x=1.1,
             y=1,

rank_data_03_25_2025.json

@@ -5,67 +5,67 @@
             {
                 "model": "gamingagent + claude-3-5-sonnet-20241022",
                 "score": 1267.7,
-                "detail_data":"709,1532,1562",
+                "detail_data": "709,1532,1562",
                 "progress": "1-1"
             },
             {
-                "model": "gamingagent + claude-3-7-sonnet-20250219 (thinking)",
+                "model": "gamingagent + claude-3-7-sonnet-20250219",
                 "score": 1418.7,
-                "detail_data":"2015,709,1532",
+                "detail_data": "2015,709,1532",
                 "progress": "1-1"
             },
             {
-                "model": "gamingagent + gemini-2.5-flash-preview-04-17 (thinking)",
+                "model": "gamingagent + gemini-2.5-flash-preview-04-17",
                 "score": 1385.0,
-                "detail_data":"1672,1266,1247",
+                "detail_data": "1672,1266,1247",
                 "progress": "1-1"
             },
             {
-                "model": "gamingagent + gemini-2.5-pro-preview-05-06 (thinking)",
+                "model": "gamingagent + gemini-2.5-pro-preview-05-06",
                 "score": 1498.3,
-                "detail_data":"1561,1271,1663",
+                "detail_data": "1561,1271,1663",
                 "progress": "1-1"
             },
             {
                 "model": "gamingagent + llama-4-maverick-17b-128e-instruct-fp8",
                 "score": 1468.7,
-                "detail_data":"898,2008,1500",
+                "detail_data": "898,2008,1500",
                 "progress": "1-1"
             },
             {
                 "model": "gamingagent + gpt-4.1-2025-04-14",
                 "score": 2126.3,
-                "detail_data":"1531,722,4126",
+                "detail_data": "1531,722,4126",
                 "progress": "1-1"
             },
             {
                 "model": "gamingagent + gpt-4o-2024-11-20",
                 "score": 2047.3,
-                "detail_data":"2017,2590,1535",
+                "detail_data": "2017,2590,1535",
                 "progress": "1-1"
             },
             {
                 "model": "gamingagent + o1-2024-12-17",
                 "score": 855,
-                "detail_data":"855",
+                "detail_data": "855",
                 "progress": "1-1"
             },
             {
                 "model": "gamingagent + o3-2025-04-16",
                 "score": 3445,
-                "detail_data":"3445",
+                "detail_data": "3445",
                 "progress": "1-1"
             },
             {
                 "model": "gamingagent + o4-mini-2025-04-16",
                 "score": 1448.0,
-                "detail_data":"1525,1263,1556",
+                "detail_data": "1525,1263,1556",
                 "progress": "1-1"
             },
             {
                 "model": "random (x30)",
                 "score": 986.97,
-                "detail_data":"986.97",
+                "detail_data": "986.97",
                 "progress": "1-1"
             }
         ]
@@ -80,7 +80,7 @@
                 "highest_tail": 256
             },
             {
-                "model": "gamingagent + claude-3-7-sonnet-20250219 (thinking)",
+                "model": "gamingagent + claude-3-7-sonnet-20250219",
                 "score": 2624,
                 "details": "2560,3224,2088",
                 "highest_tail": 256
@@ -92,19 +92,19 @@
                 "highest_tail": 256
             },
             {
-                "model": "gamingagent + gemini-2.5-flash-preview-04-17 (thinking)",
+                "model": "gamingagent + gemini-2.5-flash-preview-04-17",
                 "score": 1697.33,
                 "details": "1304,1316,2472",
                 "highest_tail": 256
             },
             {
-                "model": "gamingagent + gemini-2.5-pro-preview-05-06 (thinking)",
+                "model": "gamingagent + gemini-2.5-pro-preview-05-06",
                 "score": 3586.67,
                 "details": "5300,2400,3060",
                 "highest_tail": 512
             },
             {
-                "model": "gamingagent + grok-3-mini-beta (thinking)",
+                "model": "gamingagent + grok-3-mini-beta",
                 "score": 4036,
                 "details": "6412,2492,3204",
                 "highest_tail": 512
@@ -192,7 +192,7 @@
                 "details": "16,14,14"
             },
             {
-                "model": "gamingagent + claude-3-7-sonnet-20250219 (thinking)",
+                "model": "gamingagent + claude-3-7-sonnet-20250219",
                 "score": 16.3,
                 "details": "19,15,15"
             },
@@ -202,17 +202,17 @@
                 "details": "15,14,14"
             },
             {
-                "model": "gamingagent + gemini-2.5-flash-preview-04-17 (thinking)",
+                "model": "gamingagent + gemini-2.5-flash-preview-04-17",
                 "score": 16.3,
                 "details": "20,14,15"
             },
             {
-                "model": "gamingagent + gemini-2.5-pro-preview-05-06 (thinking)",
+                "model": "gamingagent + gemini-2.5-pro-preview-05-06",
                 "score": 23.3,
                 "details": "23,23,24"
             },
             {
-                "model": "gamingagent + grok-3-mini-beta (thinking)",
+                "model": "gamingagent + grok-3-mini-beta",
                 "score": 21.3,
                 "details": "20,15,29"
             },
@@ -287,7 +287,7 @@
                 "details": "92,165,61"
             },
             {
-                "model": "gamingagent + claude-3-7-sonnet-20250219 (thinking)",
+                "model": "gamingagent + claude-3-7-sonnet-20250219",
                 "score": 484,
                 "details": "535,428,489"
             },
@@ -297,17 +297,17 @@
                 "details": "409,436,497"
             },
             {
-                "model": "gamingagent + gemini-2.5-flash-preview-04-17 (thinking)",
+                "model": "gamingagent + gemini-2.5-flash-preview-04-17",
                 "score": 334.7,
                 "details": "259,372,373"
             },
             {
-                "model": "gamingagent + gemini-2.5-pro-preview-05-06 (thinking)",
+                "model": "gamingagent + gemini-2.5-pro-preview-05-06",
                 "score": 416.3,
                 "details": "411,414,424"
             },
             {
-                "model": "gamingagent + grok-3-mini-beta (thinking)",
+                "model": "gamingagent + grok-3-mini-beta",
                 "score": 254,
                 "details": "299,332,131"
             },
@@ -379,85 +379,85 @@
             {
                 "model": "gamingagent + claude-3-5-sonnet-20241022",
                 "score": 0,
-                "detail_box_on_target":"0,0,0",
+                "detail_box_on_target": "0,0,0",
                 "cracked_levels": "0,0,0"
             },
             {
-                "model": "gamingagent + claude-3-7-sonnet-20250219 (thinking)",
+                "model": "gamingagent + claude-3-7-sonnet-20250219",
                 "score": 2.33,
-                "detail_box_on_target":"2,4,1",
+                "detail_box_on_target": "2,4,1",
                 "cracked_levels": "1,2,0"
             },
             {
                 "model": "gamingagent + deepseek-r1-0120",
                 "score": 1.33,
-                "detail_box_on_target":"2,0,2",
+                "detail_box_on_target": "2,0,2",
                 "cracked_levels": "1,0,1"
             },
             {
-                "model": "gamingagent + gemini-2.5-flash-preview-04-17 (thinking)",
+                "model": "gamingagent + gemini-2.5-flash-preview-04-17",
                 "score": 1.67,
-                "detail_box_on_target":"3,0,2",
+                "detail_box_on_target": "3,0,2",
                 "cracked_levels": "2,0,1"
             },
             {
-                "model": "gamingagent + gemini-2.5-pro-preview-05-06 (thinking)",
+                "model": "gamingagent + gemini-2.5-pro-preview-05-06",
                 "score": 4.33,
-                "detail_box_on_target":"4,4,5",
+                "detail_box_on_target": "4,4,5",
                 "cracked_levels": "2,2,3"
             },
             {
-                "model": "gamingagent + grok-3-mini-beta (thinking)",
+                "model": "gamingagent + grok-3-mini-beta",
                 "score": 5.67,
-                "detail_box_on_target":"5,6,6",
+                "detail_box_on_target": "5,6,6",
                 "cracked_levels": "3,3,3"
             },
             {
                 "model": "gamingagent + llama-4-maverick-17b-128e-instruct-fp8",
                 "score": 0,
-                "detail_box_on_target":"0,0,0",
+                "detail_box_on_target": "0,0,0",
                 "cracked_levels": "0,0,0"
             },
             {
                 "model": "gamingagent + gpt-4.1-2025-04-14",
                 "score": 0,
-                "detail_box_on_target":"0,0,0",
+                "detail_box_on_target": "0,0,0",
                 "cracked_levels": "0,0,0"
             },
             {
                 "model": "gamingagent + gpt-4o-2024-11-20",
                 "score": 0,
-                "detail_box_on_target":"0,0,0",
+                "detail_box_on_target": "0,0,0",
                 "cracked_levels": "0,0,0"
             },
             {
                 "model": "gamingagent + o1-2024-12-17",
                 "score": 2.33,
-                "detail_box_on_target":"2,2,3",
+                "detail_box_on_target": "2,2,3",
                 "cracked_levels": "1,1,2"
             },
             {
                 "model": "gamingagent + o1-mini-2024-09-12",
                 "score": 1.33,
-                "detail_box_on_target":"1,2,1",
+                "detail_box_on_target": "1,2,1",
                 "cracked_levels": "0,1,0"
             },
             {
                 "model": "gamingagent + o3-2025-04-16",
                 "score": 8,
-                "detail_box_on_target":"10,6",
+                "detail_box_on_target": "10,6",
                 "cracked_levels": "5,3"
             },
             {
                 "model": "gamingagent + o4-mini-2025-04-16",
                 "score": 5.33,
-                "detail_box_on_target":"4,6,6",
+                "detail_box_on_target": "4,6,6",
                 "cracked_levels": "2,2,3"
             },
             {
                 "model": "random (x30)",
                 "score": 0,
-                "detail_box_on_target":"0,0,0",
+                "detail_box_on_target": "0,0,0",
                 "cracked_levels": "0,0,0"
             },
             {
@@ -492,7 +492,7 @@
                 "evaluator result": "1/3"
             },
             {
-                "model": "gamingagent + claude-3-7-sonnet-20250219 (thinking)",
+                "model": "gamingagent + claude-3-7-sonnet-20250219",
                 "score": 7,
                 "progress": "2:2/9",
                 "evaluator result": "5/11"
@@ -504,19 +504,19 @@
                 "evaluator result": "1/5"
             },
             {
-                "model": "gamingagent + gemini-2.5-flash-preview-04-17 (thinking)",
+                "model": "gamingagent + gemini-2.5-flash-preview-04-17",
                 "score": 4,
                 "progress": "1:4/5",
                 "evaluator result": "1/7"
             },
             {
-                "model": "gamingagent + gemini-2.5-pro-preview-05-06 (thinking)",
+                "model": "gamingagent + gemini-2.5-pro-preview-05-06",
                 "score": 7,
                 "progress": "2:2/9",
                 "evaluator result": "2/3"
             },
             {
-                "model": "gamingagent + grok-3-mini-beta (thinking)",
+                "model": "gamingagent + grok-3-mini-beta",
                 "score": 0,
                 "progress": "0",
                 "evaluator result": "0"

rank_single_model_03_25_2025.json

@@ -9,19 +9,19 @@
                 "progress": "1-1"
             },
             {
-                "model": "claude-3-7-sonnet-20250219 (thinking)",
+                "model": "claude-3-7-sonnet-20250219",
                 "score": 1430.0,
                 "detail_data": "1532,1515,1243",
                 "progress": "1-1"
             },
             {
-                "model": "gemini-2.5-flash-preview-04-17 (thinking)",
+                "model": "gemini-2.5-flash-preview-04-17",
                 "score": 1540.7,
                 "detail_data": "1794,1270,1558",
                 "progress": "1-1"
             },
             {
-                "model": "gemini-2.5-pro-preview-05-06 (thinking)",
+                "model": "gemini-2.5-pro-preview-05-06",
                 "score": 1025.3,
                 "detail_data": "820,1534,722",
                 "progress": "1-1"
@@ -75,67 +75,67 @@
         "results": [
             {
                 "model": "claude-3-5-sonnet-20241022",
-                "score": 17.0,
+                "score": 84,
                 "details": "188,20,44",
                 "highest_tail": 32
             },
             {
-                "model": "claude-3-7-sonnet-20250219 (thinking)",
-                "score": 126.3,
+                "model": "claude-3-7-sonnet-20250219",
+                "score": 2624,
                 "details": "1596,4256,3008",
                 "highest_tail": 512
             },
             {
-                "model": "gemini-2.5-flash-preview-04-17 (thinking)",
-                "score": 97.7,
+                "model": "gemini-2.5-flash-preview-04-17",
+                "score": 1738.67,
                 "details": "2228,1424,1564",
                 "highest_tail": 256
             },
             {
-                "model": "gemini-2.5-pro-preview-05-06 (thinking)",
+                "model": "gemini-2.5-pro-preview-05-06",
                 "score": 120.5,
-                "details": "5784,3544,3704",
+                "details": "4344",
                 "highest_tail": 512
             },
             {
                 "model": "llama-4-maverick-17b-128e-instruct-fp8",
-                "score": 44.6,
+                "score": 28,
                 "details": "16,56,12",
                 "highest_tail": 64
             },
             {
                 "model": "gpt-4.1-2025-04-14",
-                "score": 94.5,
+                "score": 1113.33,
                 "details": "264,500,2576",
                 "highest_tail": 256
             },
             {
                 "model": "gpt-4o-2024-11-20",
-                "score": 70.4,
+                "score": 176,
                 "details": "292,196,40",
                 "highest_tail": 32
             },
             {
                 "model": "o1-2024-12-17",
-                "score": 128.1,
+                "score": 7176,
                 "details": "7176",
                 "highest_tail": 512
             },
             {
                 "model": "o3-2025-04-16",
-                "score": 128.2,
+                "score": 7220,
                 "details": "7220",
                 "highest_tail": 512
             },
             {
                 "model": "o4-mini-2025-04-16",
-                "score": 97.6,
+                "score": 1882.67,
                 "details": "3004,84,2560",
                 "highest_tail": 256
             },
             {
                 "model": "random (x30)",
-                "score": 100.4,
+                "score": 1228,
                 "details": "",
                 "highest_tail": 128
             },
@@ -168,17 +168,17 @@
                 "details": "10,15,12"
             },
             {
-                "model": "claude-3-7-sonnet-20250219 (thinking)",
+                "model": "claude-3-7-sonnet-20250219",
                 "score": 13.0,
                 "details": "13,13,13"
             },
             {
-                "model": "gemini-2.5-flash-preview-04-17 (thinking)",
+                "model": "gemini-2.5-flash-preview-04-17",
                 "score": 19.0,
                 "details": "15,18,24"
             },
             {
-                "model": "gemini-2.5-pro-preview-05-06 (thinking)",
+                "model": "gemini-2.5-pro-preview-05-06",
                 "score": 12.3,
                 "details": "15,9,13"
             },
@@ -243,17 +243,17 @@
                 "details": "15,36,0"
             },
             {
-                "model": "claude-3-7-sonnet-20250219 (thinking)",
+                "model": "claude-3-7-sonnet-20250219",
                 "score": 126.3,
                 "details": "148,182,49"
             },
             {
-                "model": "gemini-2.5-flash-preview-04-17 (thinking)",
+                "model": "gemini-2.5-flash-preview-04-17",
                 "score": 97.7,
                 "details": "60,101,132"
             },
             {
-                "model": "gemini-2.5-pro-preview-05-06 (thinking)",
+                "model": "gemini-2.5-pro-preview-05-06",
                 "score": 177.3,
                 "details": "117,169,246"
             },
@@ -319,19 +319,19 @@
                 "cracked_levels": "0,0,0"
             },
             {
-                "model": "claude-3-7-sonnet-20250219 (thinking)",
+                "model": "claude-3-7-sonnet-20250219",
                 "score": 0,
                 "detail_box_on_target": "0,0,0",
                 "cracked_levels": "0,0,0"
             },
             {
-                "model": "gemini-2.5-flash-preview-04-17 (thinking)",
+                "model": "gemini-2.5-flash-preview-04-17",
                 "score": 0,
                 "detail_box_on_target": "0,0,0",
                 "cracked_levels": "0,0,0"
             },
             {
-                "model": "gemini-2.5-pro-preview-05-06 (thinking)",
+                "model": "gemini-2.5-pro-preview-05-06",
                 "score": 1,
                 "detail_box_on_target": "1,1,1",
                 "cracked_levels": "0,0,0"
@@ -407,17 +407,17 @@
                 "progress": "1:1/5"
             },
             {
-                "model": "claude-3-7-sonnet-20250219 (thinking)",
+                "model": "claude-3-7-sonnet-20250219",
                 "score": 3,
                 "progress": "1:3/5"
             },
             {
-                "model": "gemini-2.5-flash-preview-04-17 (thinking)",
+                "model": "gemini-2.5-flash-preview-04-17",
                 "score": 1,
                 "progress": "1:1/5"
             },
             {
-                "model": "gemini-2.5-pro-preview-05-06 (thinking)",
+                "model": "gemini-2.5-pro-preview-05-06",
                 "score": 8,
                 "progress": "2:3/9"
             },