mirror of
https://github.com/browser-use/browser-use.git
synced 2026-09-14 19:59:47 +08:00
Magnus Müller
145 lines
4.4 KiB
Python
145 lines
4.4 KiB
Python
import matplotlib.pyplot as plt
|
|
|
|
# Data points from the chart (time in seconds per task)
|
|
data_raw = {
|
|
'Gemini 2.5 Pro': {'time_seconds': 95, 'score': 78.7, 'company': 'Google', 'cost_per_task': 14.7},
|
|
'GPT-5': {'time_seconds': 199, 'score': 81.0, 'company': 'OpenAI', 'cost_per_task': 25.4},
|
|
'GPT-5-mini': {'time_seconds': 164, 'score': 79.0, 'company': 'OpenAI', 'cost_per_task': 4.9},
|
|
'Claude Sonnet 4.5': {'time_seconds': 186, 'score': 86.4, 'company': 'Anthropic', 'cost_per_task': 39.2},
|
|
'GPT-4.1-mini': {'time_seconds': 105, 'score': 72.8, 'company': 'OpenAI', 'cost_per_task': 4.8},
|
|
'BU 1.0': {'time_seconds': 33.4, 'score': 82.0, 'company': 'Browser-Use', 'cost_per_task': 1.9},
|
|
'GPT-4o': {'time_seconds': 123, 'score': 71.8, 'company': 'OpenAI', 'cost_per_task': 39.2},
|
|
'Gemini-Flash-Latest': {'time_seconds': 97, 'score': 80.2, 'company': 'Google', 'cost_per_task': 5.5},
|
|
}
|
|
|
|
# Convert to tasks completed with $1
|
|
# Note: cost_per_task is in cents, so convert to dollars first
|
|
data = {}
|
|
budget = 1 # dollars
|
|
for model, info in data_raw.items():
|
|
cost_per_task_dollars = info['cost_per_task'] / 100 # convert cents to dollars
|
|
tasks_with_budget = budget / cost_per_task_dollars
|
|
data[model] = {'tasks_with_100': tasks_with_budget, 'score': info['score'], 'company': info['company']}
|
|
|
|
# Company colors
|
|
colors = {
|
|
'OpenAI': '#3498db', # bright blue
|
|
'Anthropic': '#e74c3c', # bright red
|
|
'Google': '#9b59b6', # purple
|
|
'Browser-Use': '#2ecc71', # bright green
|
|
}
|
|
|
|
# Create figure with dark background
|
|
plt.figure(figsize=(12, 6))
|
|
ax = plt.gca()
|
|
ax.set_facecolor('black')
|
|
plt.gcf().patch.set_facecolor('black')
|
|
|
|
# Plot points by company
|
|
for company in ['OpenAI', 'Anthropic', 'Google', 'Browser-Use']:
|
|
company_data = {k: v for k, v in data.items() if v['company'] == company}
|
|
if company_data:
|
|
tasks = [v['tasks_with_100'] for v in company_data.values()]
|
|
scores = [v['score'] for v in company_data.values()]
|
|
plt.scatter(tasks, scores, c=colors[company], s=150, label=company, edgecolors='none', alpha=0.8, zorder=3)
|
|
|
|
# Add labels for each point
|
|
for model, info in data.items():
|
|
if model == 'BU 1.0':
|
|
offset_x = 0
|
|
offset_y = 2.5
|
|
ha = 'center'
|
|
label_text = 'BU 1.0 ★'
|
|
elif model == 'Claude Sonnet 4.5':
|
|
offset_x = 7.5
|
|
offset_y = 0
|
|
ha = 'center'
|
|
label_text = model
|
|
elif model == 'Gemini 2.5 Pro':
|
|
offset_x = 0
|
|
offset_y = -2
|
|
ha = 'center'
|
|
label_text = model
|
|
elif model == 'GPT-5-mini':
|
|
offset_x = 0
|
|
offset_y = -2
|
|
ha = 'center'
|
|
label_text = model
|
|
elif model == 'GPT-5':
|
|
offset_x = 0
|
|
offset_y = 2
|
|
ha = 'center'
|
|
label_text = model
|
|
elif model == 'GPT-4.1-mini':
|
|
offset_x = 0
|
|
offset_y = -2
|
|
ha = 'center'
|
|
label_text = model
|
|
elif model == 'GPT-4o':
|
|
offset_x = 0
|
|
offset_y = -2
|
|
ha = 'center'
|
|
label_text = model
|
|
elif model == 'Gemini-Flash-Latest':
|
|
offset_x = 0
|
|
offset_y = 2
|
|
ha = 'center'
|
|
label_text = model
|
|
else:
|
|
offset_x = 0
|
|
offset_y = 2.5
|
|
ha = 'center'
|
|
label_text = model
|
|
|
|
plt.text(
|
|
info['tasks_with_100'] + offset_x,
|
|
info['score'] + offset_y,
|
|
label_text,
|
|
fontsize=14,
|
|
color='white',
|
|
ha=ha,
|
|
va='center',
|
|
fontweight='bold',
|
|
)
|
|
|
|
# Draw dashed lines connecting pareto frontier (upper-right is optimal: high tasks, high score)
|
|
pareto_points = [
|
|
('BU 1.0', 'Claude Sonnet 4.5'),
|
|
]
|
|
|
|
for p1, p2 in pareto_points:
|
|
x = [data[p1]['tasks_with_100'], data[p2]['tasks_with_100']]
|
|
y = [data[p1]['score'], data[p2]['score']]
|
|
plt.plot(x, y, 'k--', linewidth=1, alpha=0.5, zorder=1)
|
|
|
|
# Styling
|
|
plt.xlabel('Tasks completed with $1', fontsize=14, color='white', fontweight='bold')
|
|
plt.ylabel('Accuracy', fontsize=14, color='white', fontweight='bold')
|
|
plt.title('Performance on WebBench-200', fontsize=14, color='white', pad=20)
|
|
|
|
# Set axis limits - calculate based on data range
|
|
max_tasks = max(v['tasks_with_100'] for v in data.values())
|
|
plt.xlim(0, max_tasks * 1.1)
|
|
plt.ylim(60, 90)
|
|
|
|
# Grid
|
|
plt.grid(True, linestyle='--', alpha=0.3, color='gray')
|
|
|
|
# Legend
|
|
plt.legend(loc='lower left', ncol=2, framealpha=0.9, facecolor='black', edgecolor='white', fontsize=10, labelcolor='white')
|
|
|
|
# Tick colors
|
|
ax.tick_params(colors='white', which='both')
|
|
for spine in ax.spines.values():
|
|
spine.set_color('white')
|
|
|
|
# Make tick labels bold
|
|
for label in ax.get_xticklabels():
|
|
label.set_fontweight('bold')
|
|
for label in ax.get_yticklabels():
|
|
label.set_fontweight('bold')
|
|
|
|
plt.tight_layout()
|
|
plt.savefig('webbench_performance_cost.png', dpi=300, facecolor='black')
|
|
plt.close()
|