Estimate what an AI feature will cost each month before you build it. Enter your traffic and prompt size, and the calculator compares current list prices for OpenAI, Anthropic, Google and DeepSeek models.
How to estimate your inputs
- Requests per day. Count every model call, not every user. An agent that takes five steps to answer one question makes five requests.
- Input tokens. Your system prompt, instructions, conversation history and any retrieved documents, all sent again on every request. About 750 English words is 1,000 tokens. Long chat histories grow this number quickly.
- Output tokens. The response length, plus any reasoning or “thinking” tokens on reasoning models, which are billed as output.
- Cache share. Most providers charge much less for input that repeats exactly from a recent request, such as a long fixed system prompt. If your prompt starts with the same long block every time, 50% to 80% caching is realistic.
What the numbers show
With the default inputs (1,000 requests a day, 2,000 input and 500 output tokens, no caching), the monthly bill ranges from about $14 on the smallest models (GPT-6 Luna and Claude Haiku 5.5) to $1,350 on the largest (GPT-6 Astra and Claude Fable 5.1). That spread is the point: for many features, such as classification, extraction or short summaries, a small model does the job at a fraction of the cost. Test the cheapest model that might work first, and move up only where quality falls short.
Output tokens usually cost four to five times as much as input tokens, so asking for shorter answers is one of the cheapest optimizations you can make.
Prices in the calculator
| Model | Provider | Input / 1M tokens | Output / 1M tokens | Cached input | Notes |
|---|---|---|---|---|---|
| GPT-6 Astra | OpenAI | $10.00 | $50.00 | $1.000 | Short-context standard price; long context costs more. |
| GPT-6.1 Sol | OpenAI | $2.00 | $10.00 | $0.100 | Short-context standard price. |
| GPT-6.1 Terra | OpenAI | $1.00 | $5.00 | $0.050 | Short-context standard price. |
| GPT-6 Luna | OpenAI | $0.100 | $0.500 | $0.010 | Short-context standard price. |
| Claude Fable 5.1 | Anthropic | $10.00 | $50.00 | $0.250 | |
| Claude Opus 5.5 | Anthropic | $4.00 | $20.00 | $0.200 | |
| Claude Sonnet 5.5 | Anthropic | $2.00 | $10.00 | $0.100 | |
| Claude Haiku 5.5 | Anthropic | $0.100 | $0.500 | $0.010 | Prompts up to 100,000 tokens; longer prompts cost $0.50 in / $2.50 out. |
| Gemini 4 Argon | $2.00 | $10.00 | – | Price through Nov 30, 2026; Google lists $4.00 in / $20.00 out from Dec 1, 2026. Output includes thinking tokens. | |
| Gemini 3.8 Flash | $0.750 | $3.75 | – | Price through Dec 31, 2026; Google lists $1.50 in / $7.50 out from Jan 1, 2027. | |
| Gemini 3.1 Flash-Lite | $0.250 | $1.50 | – | Text, image and video input; audio input costs more. | |
| DeepSeek Flash (V4.1) | DeepSeek | $0.300 | $1.20 | $0.006 | Peak-hour price (cache miss). Off-peak is half. |
| DeepSeek V4 Pro | DeepSeek | $1.32 | $3.96 | $0.044 | Peak-hour price (cache miss). Off-peak is half. |
Official list prices checked Oct 30, 2026, before tax. Sources: platform.openai.com, docs.claude.com, ai.google.dev, api-docs.deepseek.com.
Announced price changes: Google lists Gemini 4 Argon rising from $2/$10 to $4/$20 per million input/output tokens on December 1, 2026, and Gemini 3.8 Flash rising from $0.75/$3.75 to $1.50/$7.50 on January 1, 2027. DeepSeek charges half its listed price during off-peak hours.
For a model-by-model breakdown, read AI API pricing compared.
Frequently asked questions
How many tokens is a word?
In English, one token is about three quarters of a word, so 1,000 tokens is roughly 750 words. Code and other languages often use more tokens per word.
Are reasoning tokens billed?
Yes. Reasoning models produce hidden thinking tokens before the answer, and providers bill them as output tokens. Include them in the output estimate; they can be several times longer than the visible answer.
Is a ChatGPT or Claude subscription the same as API access?
No. Consumer plans such as ChatGPT Plus and Claude Pro are billed separately from the API. API usage is pay per token through the provider’s developer platform.
How can I lower my API bill?
Use the smallest model that meets your quality bar, cache long fixed prompts, cap output length, trim conversation history, and use batch processing for work that does not need an instant answer.