final udpates

app.py

@@ -1,7 +1,6 @@
 import gradio as gr
 import pandas as pd
 import plotly.graph_objects as go
-import plotly.express as px
 import numpy as np
 from model_handler import (
     ModelHandler,
@@ -75,22 +74,28 @@ def create_leaderboard_chart(leaderboard_df):
     if leaderboard_df.empty or "Average" not in leaderboard_df.columns:
         return None
 
-    # Get top 20 models (reverse for display with best at top)
-    top_20 = leaderboard_df.head(20).copy().iloc[::-1]
-    colors = ["#F4DD2C" if x == '-' else "#7DCEF1" for x in top_20['Size']]
-
-    # Create horizontal bar chart
-    fig = go.Figure(data=[
-        go.Bar(
-            y=top_20["Model"],
-            x=top_20["Average"],
+    # Get top 20 models sorted by average, best at top in horizontal bar
+    top_20 = leaderboard_df.head(20).copy().sort_values("Average", ascending=True)
+    # Split into two traces for legend
+    closed_mask = top_20['Size'] == '-'
+    open_mask = ~closed_mask
+
+    fig = go.Figure()
+    for mask, label, color in [
+        (closed_mask, "Closed Source", "#F4DD2C"),
+        (open_mask, "Open Source", "#7DCEF1"),
+    ]:
+        subset = top_20[mask]
+        fig.add_trace(go.Bar(
+            y=subset["Model"],
+            x=subset["Average"],
             orientation="h",
-            marker_color=colors,
-            text=top_20["Average"].round(4),
+            name=label,
+            marker_color=color,
+            text=subset["Average"].round(4),
             textposition="outside",
             hovertemplate="<b>%{y}</b><br>Avg Score: %{x:.4f}<extra></extra>",
-        )
-    ])
+        ))
 
     fig.update_layout(
         title="Top 20 Models by Average Score",
@@ -99,6 +104,8 @@ def create_leaderboard_chart(leaderboard_df):
         xaxis_range  = [0,1],
         height=600,
         hovermode="y unified",
+        showlegend=True,
+        legend=dict(x=1, y=1, xanchor="right", yanchor="top", bgcolor="rgba(0,0,0,0)", font=dict(color="white")),
         margin=dict(l=100),
         paper_bgcolor="rgba(30,30,30,0.85)",
         plot_bgcolor="rgba(30,30,30,0.85)",
@@ -114,6 +121,8 @@ def create_leaderboard_chart(leaderboard_df):
             dtick=1,
             tickfont=dict(color="white"),
             title_font=dict(color="white"),
+            categoryorder="array",
+            categoryarray=top_20["Model"].tolist(),
         )
     )
 
@@ -175,28 +184,28 @@ def create_spend_report_chart(leaderboard_df):
     #     font=dict(size=11),
     #     margin=dict(l=80, r=80, t=80, b=80),
     # )
-    # === Small jitter so points don't land exactly on top of each other ===
-    merged["Cost_jitter"] = merged["Total Cost ($)"] + np.random.normal(0, 0.8, len(merged))
     merged["Model"] = merged["Model"].str.split("/").str[-1]
-    fig = px.scatter(
-        merged,
-        x="Total Cost ($)",           # jittered X so points separate
-        y="Average",
-        text="Model",              # ← model name appears on the plot
-        # color="Average",     # color by accuracy (green = better)
-        # color_continuous_scale=px.colors.sequential.Viridis,
-        hover_name="Model",
-        hover_data=["Total Cost ($)", "Total Tokens"],
-        title="Model Evaluation: Cost vs Accuracy<br><sub>Hover or read labels for full details</sub>",
-        labels={"Average": "Accuracy (Average Score)"},
-        log_x=True,
-    )
 
-    fig.update_traces(
-        textposition="middle right",     # labels sit above each point
-        textfont=dict(size=11, color="white"),
-        marker=dict(size=14, line=dict(width=1, color="white")),  # bigger, visible points
-    )
+    fig = go.Figure()
+    for mask, label, color in [
+        (merged['Size'] == '-', "Closed Source", "#F4DD2C"),
+        (merged['Size'] != '-', "Open Source", "#7DCEF1"),
+    ]:
+        subset = merged[mask]
+        print('[MODELS]',subset['Model'].tolist())
+        fig.add_trace(go.Scatter(
+            x=subset["Total Cost ($)"],
+            y=subset["Average"],
+            mode="markers+text",
+            name=label,
+            marker=dict(size=14, color=color, line=dict(width=1, color="white")),
+            text=subset["Model"],
+            textposition=["top right","middle right","middle right","middle right","top right","middle right","top right","middle right","bottom left","middle left","middle left","middle left","middle right"] if len(subset)>2 else ['bottom right','middle right'],
+            textfont=dict(size=11, color="white"),
+            hovertemplate="<b>%{text}</b><br>Cost: $%{x:.2f}<br>Avg Score: %{y:.4f}<extra></extra>",
+        ))
+
+    fig.update_layout(title="Model Evaluation: Cost vs Accuracy<br><sub>Hover or read labels for full details</sub>")
 
     fig.update_layout(
         paper_bgcolor="rgba(30,30,30,0.85)",
@@ -205,13 +214,16 @@ def create_spend_report_chart(leaderboard_df):
         height=650,
         xaxis_title="Total Cost ($)",
         yaxis=dict(range=[0, 1], showgrid=False),
-        xaxis=dict(showgrid=False),
-        showlegend=False,
+        # xaxis=dict(showgrid=False, dtick=10, tickprefix="$"),
+        showlegend=True,
+        legend=dict(x=1, y=1, xanchor="right", yanchor="top", bgcolor="rgba(0,0,0,0)", font=dict(color="white")),
         margin=dict(l=40, r=40, t=80, b=40)
     )
-
-    # Optional: show the real cost value on the x-axis (not the jittered one)
-    fig.update_xaxes(tickvals=merged["Total Cost ($)"], ticktext=[f"${v:.1f}" for v in merged["Total Cost ($)"]])
+    fig.update_xaxes(
+        showgrid=False,
+        tickprefix="$",
+        tickmode="auto",
+        )
 
     return fig
 
@@ -337,7 +349,7 @@ def refresh_data():
         "detailed": prepare_detailed_leaderboard(detailed_results, leaderboard),
         "task_scores": task_scores,
     }
-
+    global_data["leaderboard"].to_csv('leaderboard.csv',index =False)
     return (
         create_leaderboard_chart(global_data["leaderboard"]),
         global_data["leaderboard"],
@@ -398,7 +410,7 @@ def main():
                 )
 
                 gr.Markdown(
-                    "> **Note:** We observed significant reliability issues with Claude 4.6 (same for 4.5) during evaluation. "
+                    "> **Note:** We observed significant reliability issues with Claude 4.6 (4.5) during evaluation. "
                     "To preserve the integrity of the benchmark, all Claude 4.6 (4.5) models have been excluded from the leaderboard. "
                     "You can learn more about the issue in [this GitHub issue](https://github.com/anthropics/claude-code/issues/17357)."
                 )
@@ -425,7 +437,7 @@ def main():
 
                     **Notes:**
                     - Token counts are approximations based on API usage patterns
-                    - Spend data is only reported for API-based evaluations; other models were evaluated on local compute and costs are not reported
+                    - Spend data is only reported for models having more than 50B parameters.
                     """
                 )
 

full_price_report.csv

@@ -13,5 +13,4 @@ openai/gpt-5.4-nano,1105571,0.623154
 x-ai/grok-4-fast,10881722,2.711288
 x-ai/grok-4.1-fast,12961678,3.236579
 x-ai/grok-4.20-beta,1554643,2.749659
-z-ai/glm-4.7-flash,15109580,2.976598
 z-ai/glm-5,17845348,26.480097