Cost Estimator
AI API cost calculator online. Estimate costs for GPT-4, Claude, Gemini APIs. Input and output token pricing—free LLM cost tool.
openai Input: $2.5/1M | Output: $15/1M
Quick presets:
Input Cost
$0.002500
1,000 tokens
Output Cost
$0.007500
500 tokens
Total per Request
$0.0100
1,500 tokens
Compare Models (select up to 3)
⚠️
Approximate Pricing
Prices last updated: 2026-07-08. Always check provider websites for current rates.
Actual costs may vary based on region, volume discounts, and API version.
Quick Monthly Estimates
100/day
$30.00
/month
1,000/day
$300.00
/month
10K/day
$3000.00
/month
100K/day
$30000.00
/month
Features
- Highly accurate, up-to-date pricing algorithms for top-tier LLM providers (OpenAI, Anthropic, Google, Meta, Mistral)
- Distinct calculation streams for Input (Prompt) tokens versus Output (Completion) tokens
- Advanced support for calculating Prompt Caching discounts (e.g., Anthropic Claude 3.5 cached input rates)
- Dynamic monthly SaaS budgeting based on projected Daily Active Users (DAU) and batch API scaling
- Direct pasting support: paste raw text to instantly calculate both the token count and the exact monetary cost
- Interactive side-by-side model comparison to instantly identify the most cost-effective AI engine
Common Use Cases
- Pitching AI feature integrations to stakeholders by providing exact monthly API budget forecasts
- Performing a rigorous cost-benefit analysis between using a flagship model (GPT-4o) versus a fast model (GPT-4o-mini)
- Calculating the exact monetary cost of processing a massive 1-million-token RAG context window
- Estimating the overhead of "invisible" reasoning tokens generated by models like OpenAI o1 or DeepSeek R1
- Budgeting for massive offline batch processing jobs (which often receive a 50% API discount)
The Asymmetric Economics of LLM APIs
Generative AI pricing is almost universally calculated dynamically based on volume, specifically measured per 1 million tokens ($/1M).
The Core Pricing Dynamics:
- Asymmetric Billing (Input vs Output): The computational power required to read text (Input) is drastically lower than the power required to generate new text (Output). Therefore, Output tokens are typically priced 3x to 5x higher than Input tokens.
- The Intelligence Premium: Frontier, reasoning-capable models (like Claude 3.5 Sonnet or GPT-4o) command an immense premium—often costing 20x to 50x more per token than their smaller, faster counterparts (like Claude 3.5 Haiku or GPT-4o-mini).
- Prompt Caching Discounts: Modern APIs now offer massive discounts (often 50% to 90% off) for Cached Input. If you repeatedly send the exact same massive system prompt or RAG document within a short time window, the provider caches the initial computation, drastically dropping your bill.
Examples
Valid - Standard Chat Query
Input: 500 tokens (System + History)
Output: 200 tokens
Cost Profile: Negligible. Best handled by fast, cheap models. Valid - Massive RAG Query
Input: 100,000 tokens (10 PDF Documents)
Output: 500 tokens
Cost Profile: Highly asymmetric. The massive input dominates the cost unless Prompt Caching is utilized. Valid - Reasoning Model Generation
Input: 1,000 tokens
Output: 5,000 tokens (4,500 invisible reasoning tokens + 500 visible answer tokens)
Cost Profile: Extremely high. You pay for the "thinking" time.Frequently Asked Questions
How frequently do LLM API prices change?
The AI industry is currently in a massive "race to the bottom". Providers like OpenAI, Google, and DeepSeek frequently slash prices (sometimes by 50% overnight) to undercut competitors. This calculator uses the most recently published public API rates.
What is the difference between a "Prompt" and a "Completion"?
These are the legacy terms used by OpenAI. The "Prompt" is your Input—the text, history, and instructions you send to the server. The "Completion" is the Output—the novel text the AI model generates and sends back to you.
How are reasoning tokens (like in OpenAI o1 or DeepSeek R1) billed?
Models equipped with Chain-of-Thought (CoT) reasoning generate thousands of internal, invisible tokens before they output the final answer. Providers bill these invisible reasoning tokens at the exact same high rate as standard Output tokens, making reasoning models significantly more expensive to run in production.
What are the best strategies to reduce my monthly AI API bill?
1. Model Routing: Use cheap models (like Llama 3 8B or GPT-4o-mini) for 80% of simple tasks, and route only the hardest 20% to flagship models. 2. Implement Semantic Caching to avoid querying the API for identical user questions. 3. Utilize the Batch API for non-urgent background tasks (which guarantees a 50% discount).
💡 Tips
- Always calculate your Input (Prompt) and Output (Generation) estimates separately. RAG apps are entirely Input-heavy, while creative writing apps are entirely Output-heavy.
- Do not forget to multiply your estimates by your expected Daily Active Users (DAU) and add a 15% buffer for retries, errors, and system prompt overhead.