Input price
This is the number your unit economics are built on — understand it once and price anything
What is input price
Input price is what you pay to send text to a model.
Every time your application sends a prompt — a question, an instruction, a document, a conversation history — those words become tokens, and you pay for each one. The input price is the cost per million of those tokens.
That is it. Text in, cost calculated.
The number that makes it real
On sourc.dev, input prices currently range from $0.03/1M tokens at the low end to $15.00/1M at the high end. Claude 3.5 Sonnet sits at $3.00/1M. GPT-4o at $5.00/1M. DeepSeek V3 at $0.27/1M. Verified March 2026.
A tenfold difference between models is common. Understanding input price is how you decide whether that difference matters for your use case.
Why this matters to you
Input price is the number you multiply everything else against.
Estimated monthly volume: 50 million tokens of input. At $0.27/1M: $13.50/month. At $3.00/1M: $150/month. At $15.00/1M: $750/month.
Same volume. Same product. $13.50 vs $750. The model choice made that difference, not the engineering.
Knowing your input price — and your estimated monthly token volume — turns a vague "this might get expensive" into a number you can plan around.
How to calculate yours
1. Estimate your average prompt length in words. Divide by 0.75 to get tokens. 2. Add your system prompt token count — it repeats on every call. 3. Multiply by your expected monthly call volume. 4. Multiply by the model's input price per million. Divide by one million.
That is your monthly input cost. Run it against three models before you commit to one.
The part people miss
Input price is only half the story. Output price — what you pay for the model's response — is almost always higher. For most models, output costs 3–5× more per token than input. If your use case generates long responses, output price matters more than input price.
See [Output price](/glossary/output-price) for the other half.
Verified March 2026 · Source: Anthropic, OpenAI, Google, DeepSeek pricing pages