cost-analyze

/cost-analyze

Analyze and optimize your LLM API costs with current 2026 pricing.

What I Need

To analyze your costs, provide:

1. Which models are you using? (Claude, GPT-4, etc.)
2. Approximate monthly usage (requests or tokens)
3. Your current monthly bill (optional, for validation)

Or share your code that calls the LLM API and I'll estimate costs.

Current Pricing (2026)

Anthropic Claude

Model	Input (per 1M tokens)	Output (per 1M tokens)
Claude Opus 4.5	$15.00	$75.00
Claude Sonnet 4	$3.00	$15.00
Claude Haiku 3.5	$0.80	$4.00

Prompt Caching Discounts:

• Cache write: 25% more than base
• Cache read: 90% discount (10% of base price)
• 5-minute TTL

OpenAI

Model	Input (per 1M tokens)	Output (per 1M tokens)
GPT-4o	$2.50	$10.00
GPT-4o-mini	$0.15	$0.60
o1	$15.00	$60.00
o1-mini	$3.00	$12.00

Cost Analysis Workflow

Step 1: Calculate Current Costs

// Example calculation
const usage = {
  model: 'claude-3.5-sonnet',
  requestsPerDay: 1000,
  avgInputTokens: 2000,
  avgOutputTokens: 500
};

const monthlyCost = calculateMonthlyCost(usage);
// Input: 1000 * 2000 * 30 = 60M tokens/month
// Output: 1000 * 500 * 30 = 15M tokens/month
// Cost: (60 * $3) + (15 * $15) = $180 + $225 = $405/month

Step 2: Identify Optimization Opportunities

Opportunity	Potential Savings
Prompt Caching	50-90% on repeated prompts
Model Routing	40-70% using smaller models for simple tasks
Prompt Optimization	20-40% by reducing token usage
Batching	50% with Batch API (24hr turnaround)

Step 3: Implement Optimizations

A. Prompt Caching (Anthropic)

// Before: $3.00/M input tokens
const response = await anthropic.messages.create({
  model: 'claude-3.5-sonnet',
  system: longSystemPrompt,  // 4000 tokens, repeated every call
  messages: [{ role: 'user', content: userMessage }]
});

// After: $0.30/M for cached tokens (90% savings)
const response = await anthropic.messages.create({
  model: 'claude-3.5-sonnet',
  system: [
    {
      type: 'text',
      text: longSystemPrompt,
      cache_control: { type: 'ephemeral' }  // Cache this
    }
  ],
  messages: [{ role: 'user', content: userMessage }]
});