1 files changed, 159 insertions, 0 deletions
diff --git a/notebooks/llm-stats.py b/notebooks/llm-stats.py
new file mode 100644
index 0000000..cf5b9fc
--- /dev/null
+++ b/notebooks/llm-stats.py
@@ -0,0 +1,159 @@
+import marimo
+
+__generated_with = "0.23.10"
+app = marimo.App(width="full")
+
+
+@app.cell
+def _():
+    import marimo as mo
+    import pandas as pd
+    import altair as alt
+
+    return alt, mo, pd
+
+
+@app.cell
+def _(pd):
+    # Data — add new models here.
+    #   price: blended $/1M tokens
+    # https://artificialanalysis.ai/leaderboards/models
+    MODELS = [
+        {
+            "name": "DeepSeek V4 Flash",
+            "aai_score": 37,
+            "price": 0.08,
+            "tok_per_sec": 101,
+        },
+        {
+            "name": "DeepSeek V4 Pro",
+            "aai_score": 44,
+            "price": 0.18,
+            "tok_per_sec": 85,
+        },
+        {
+            "name": "Claude Opus 4.7",
+            "aai_score": 54,
+            "price": 3.85,
+            "tok_per_sec": 51,
+        },
+        {
+            "name": "Claude Opus 4.8",
+            "aai_score": 56,
+            "price": 3.85,
+            "tok_per_sec": 58,
+        },
+        {
+            "name": "Claude Sonnet 4.6",
+            "aai_score": 47,
+            "price": 2.31,
+            "tok_per_sec": 50,
+        },
+        {
+            "name": "Claude Haiku 4.5",
+            "aai_score": 30,
+            "price": 0.77,
+            "tok_per_sec": 93,
+        },
+        {
+            "name": "Gemini 3.5 Flash",
+            "aai_score": 50,
+            "price": 1.31,
+            "tok_per_sec": 161,
+        },
+        {
+            "name": "MiMo 2.5",
+            "aai_score": 40,
+            "price": 0.06,
+            "tok_per_sec": 79,
+        },
+        {
+            "name": "Nemotron 3 Ultra",
+            "aai_score": 38,
+            "price": 0.58,
+            "tok_per_sec": 164,
+        },
+        {
+            "name": "Gemma 4 E2B (local)",
+            "aai_score": 9,
+            "price": 0.0,
+            "tok_per_sec": 60,
+        },
+        {
+            "name": "GLM 5.2",
+            "aai_score": 51,
+            "price": 0.90,
+            "tok_per_sec": 82
+        }
+    ]
+
+    df = pd.DataFrame(MODELS)
+    return MODELS, df
+
+
+@app.cell
+def _(MODELS, mo):
+    mo.md(f"""
+    # 🤖 LLM Stats Tracker
+
+    **{len(MODELS)} models tracked** — edit the `MODELS` list above to add more.
+    """)
+    return
+
+
+@app.cell
+def _(alt, df):
+    # Price vs Quality (colour = tok/sec)
+    scatter = (
+        alt.Chart(df)
+        .mark_circle(size=200)
+        .encode(
+            x=alt.X("price", title="Blended Price $/1M"),
+            y=alt.Y("aai_score", title="AAI Score"),
+            color=alt.Color("tok_per_sec:Q", title="tok/sec", scale=alt.Scale(scheme="viridis")),
+            tooltip=["name", "aai_score", "price", "tok_per_sec"],
+        )
+    )
+    labels = (
+        alt.Chart(df)
+        .mark_text(dy=-12, fontSize=10)
+        .encode(
+            x="price",
+            y="aai_score",
+            text="name",
+        )
+    )
+    chart_price_quality = (scatter + labels).properties(width="container")
+    chart_price_quality
+    return
+
+
+@app.cell
+def _(alt, df):
+    # Speed vs Quality (colour = price)
+    bubble = (
+        alt.Chart(df)
+        .mark_circle(size=120, opacity=0.7)
+        .encode(
+            x=alt.X("tok_per_sec:Q", title="Tokens/sec"),
+            y=alt.Y("aai_score:Q", title="AAI Score"),
+            color=alt.Color("price:Q", title="Price $/1M", scale=alt.Scale(scheme="viridis")),
+            tooltip=["name", "aai_score", "price", "tok_per_sec"],
+        )
+    )
+    bubble_labels = (
+        alt.Chart(df)
+        .mark_text(dy=-14, fontSize=10, fontWeight="bold")
+        .encode(
+            x="tok_per_sec:Q",
+            y="aai_score:Q",
+            text="name:N",
+        )
+    )
+    chart_speed_quality = (bubble + bubble_labels).properties(width="container")
+    chart_speed_quality
+    return
+
+
+if __name__ == "__main__":
+    app.run()