AI API Pricing Comparison 2026 - GPT-5, Claude, Gemini, Llama Complete Guide

# Example: Processing a 50K token document

# Option 1: Gemini 2.5 Pro (2M context)
# Can process entire document in one call
cost_gemini = (50000 * 1.25 + 1000 * 5.00) / 1_000_000
# = $0.0675

# Option 2: GPT-5 (128K context)
# Must split into chunks and combine
chunks = 50000 / 100000  # Need 1 chunk
cost_gpt5 = (50000 * 5.00 + 1000 * 25.00) / 1_000_000
# = $0.275

# Gemini saves 75%!

# With 10% failure rate and 3 retries
base_cost = 1000  # $1000/month
retry_cost = base_cost * 0.10 * 3  # 10% fail, 3 retries each
total_cost = base_cost + retry_cost  # $1300/month

# Hidden cost: $300/month (30% increase)

def route_to_model(task_complexity, budget):
    if task_complexity == "simple":
        return "deepseek-chat"  # $0.21/1M
    elif task_complexity == "medium":
        if budget == "high":
            return "claude-sonnet-4.5"  # $4.50/1M
        else:
            return "gemini-2.5-pro"  # $3.13/1M
    else:  # complex
        if budget == "high":
            return "claude-opus-4.5"  # $22.50/1M
        else:
            return "gpt-5"  # $15.00/1M

# Example savings
# 70% simple tasks: deepseek ($0.21)
# 20% medium tasks: gemini ($3.13)
# 10% complex tasks: claude-opus ($22.50)
# Weighted average: $3.00/1M vs $15.00/1M for gpt-5
# Savings: 80%

# Without caching
monthly_requests = 100000
cost_per_request = 0.01
total_cost = monthly_requests * cost_per_request
# = $1000/month

# With 50% cache hit rate
cached_requests = monthly_requests * 0.50
api_requests = monthly_requests * 0.50
total_cost = api_requests * cost_per_request
# = $500/month

# Savings: 50%

# Individual requests
requests = 1000
cost_per_request = 0.01
total = requests * cost_per_request
# = $10

# Batch of 10 items per request
batched_requests = 1000 / 10
cost_per_batch = 0.015  # Slightly more per request
total = batched_requests * cost_per_batch
# = $1.50

# Savings: 85%

from openai import OpenAI

client = OpenAI(
    api_key="sk-your-api-key",
    base_url="https://crazyrouter.com/v1"
)

# Free model!
response = client.chat.completions.create(
    model="gemini-2.0-flash-exp",
    messages=[{"role": "user", "content": "Hello!"}]
)

# Cost: $0.00

def calculate_savings(current_model, new_model, monthly_tokens):
    """Calculate monthly savings by switching models"""

    pricing = {
        "gpt-5": 15.00,
        "claude-opus-4.5": 22.50,
        "claude-sonnet-4.5": 4.50,
        "gemini-2.5-pro": 3.13,
        "llama-3.3-70b": 0.60,
        "deepseek-chat": 0.21
    }

    current_cost = (monthly_tokens / 1_000_000) * pricing[current_model]
    new_cost = (monthly_tokens / 1_000_000) * pricing[new_model]
    savings = current_cost - new_cost
    savings_percent = (savings / current_cost) * 100

    return {
        "current_cost": current_cost,
        "new_cost": new_cost,
        "monthly_savings": savings,
        "annual_savings": savings * 12,
        "savings_percent": savings_percent
    }

# Example: 100M tokens/month
result = calculate_savings("gpt-5", "deepseek-chat", 100_000_000)

print(f"Current cost: ${result['current_cost']:,.2f}/month")
print(f"New cost: ${result['new_cost']:,.2f}/month")
print(f"Monthly savings: ${result['monthly_savings']:,.2f}")
print(f"Annual savings: ${result['annual_savings']:,.2f}")
print(f"Savings: {result['savings_percent']:.1f}%")

# Output:
# Current cost: $1,500.00/month
# New cost: $21.00/month
# Monthly savings: $1,479.00
# Annual savings: $17,748.00
# Savings: 98.6%

# Track your usage for 1 week
total_tokens = 0
total_cost = 0

for request in your_api_calls:
    tokens = request.usage.total_tokens
    cost = calculate_cost(request.model, tokens)
    total_tokens += tokens
    total_cost += cost

monthly_projection = total_cost * 4
print(f"Projected monthly cost: ${monthly_projection:,.2f}")

# Test 3 models with same prompts
models = ["gpt-5", "claude-sonnet-4.5", "deepseek-chat"]
test_prompts = [...]  # Your typical prompts

for model in models:
    quality_score = test_model(model, test_prompts)
    cost = calculate_cost(model, test_prompts)
    print(f"{model}: Quality={quality_score}, Cost=${cost}")