Latency percentiles per model PRO

preloaded fixtures

import pandas as pd
import numpy as np

requests = pd.DataFrame({
    "org": ["Northwind", "Northwind", "Playform", "Serene", "Hooli", "Hooli", "Globex", "Globex"],
    "model": ["gpt-flash", "gpt-pro", "gpt-flash", "gpt-flash", "gpt-pro", "gpt-flash", "gpt-pro", "gpt-flash"],
    "input_tokens": [1200, 5000, 300, 150, 12000, 800, 7000, 1100],
    "output_tokens": [400, 1800, 90, 40, 4200, 200, 2500, 350],
    "latency_ms": [820, 3200, 450, 510, 5100, 530, 3900, 700],
    "status": ["ok", "ok", "ok", "error", "ok", "ok", "ok", "rate_limited"],
})

pricing = pd.DataFrame({
    "model": ["gpt-mini", "gpt-flash", "gpt-pro"],
    "input_per_1k": [0.0002, 0.0010, 0.0100],
    "output_per_1k": [0.0008, 0.0030, 0.0300],
})