data update

data/csv/models_data.csv

@@ -18,7 +18,7 @@ T,Model,b4bqa,b4b,medmcqa_g2b,medmcqa_orig_filtered,medmcqa_diff,medqa_4options_
 🔶,"<a target=""_blank"" href=""https://huggingface.co/ProbeMedicalYonseiMAILab/medllama3-v20"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ProbeMedicalYonseiMAILab-medllama3-v20</a>",71.93,74.75,65.23,80.17,-14.94,76.46,90.21,-13.75
 🟢,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen2-72B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Qwen-Qwen2-72B</a>",91.02,83.72,71.55,77.87,-6.32,74.07,75.4,-1.33
 🟢,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen2-7B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Qwen-Qwen2-7B</a>",80.41,70.28,55.17,63.51,-8.34,53.7,58.99,-5.29
-💬,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen2-7B-v2"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">GPT-4</a>",94.92,,88.79,91.67,-2.88,89.95,92.33,-2.38
-💬,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen2-7B-v2"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">GPT-4o</a>",96.48,,86.49,90.52,-4.03,88.36,90.21,-1.85
-💬,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen2-7B-v2"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">GPT-3.5 Turbo</a>",91.74,,97.7,98.28,-0.58,96.03,96.3,-0.27
-💬,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen2-7B-v2"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Claude Opus</a>",92.19,,79.89,86.49,-6.6,83.33,85.71,-2.38
+💬,"<a target=""_blank"" href=""https://platform.openai.com/docs/models"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">GPT-4</a>",94.92,,88.79,91.67,-2.88,89.95,92.33,-2.38
+💬,"<a target=""_blank"" href=""https://platform.openai.com/docs/models"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">GPT-4o</a>",96.48,,86.49,90.52,-4.03,88.36,90.21,-1.85
+💬,"<a target=""_blank"" href=""https://platform.openai.com/docs/models"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">GPT-3.5 Turbo</a>",91.74,,97.7,98.28,-0.58,96.03,96.3,-0.27
+💬,"<a target=""_blank"" href=""https://www.anthropic.com/api"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Claude Opus</a>",92.19,,79.89,86.49,-6.6,83.33,85.71,-2.38