LLM API cost: what one AI feature costs per month.
LLM API cost is tokens multiplied by price. In our example, a feature with 10,000 requests a month, using 2,000 input and 500 output tokens each, costs about $4.50 on the cheapest models we priced and $90 on a mainstream one. Those are list prices checked on 5 November 2026, and your own usage will change the numbers.
Reviewed by Anubhav Gupta, Bespoke Devs · Updated
The sum
Monthly cost = (input tokens ÷ 1,000,000 × input price) + (output tokens ÷ 1,000,000 × output price).
Our example uses made-up numbers: 10,000 requests a month, with 2,000 input tokens and 500 output tokens per request. That’s 20 million input tokens and 5 million output tokens.
LLM API cost comparison for that example
| Model | Input per 1M | Output per 1M | Example month |
|---|---|---|---|
| Claude Haiku 5.5 (prompts up to 100,000 tokens) | $0.10 | $0.50 | $4.50 |
| OpenAI gpt-6-luna | $0.10 | $0.50 | $4.50 |
| Gemini gemini-3.1-flash-lite | $0.25 | $1.50 | $12.50 |
| Gemini gemini-3.8-flash, through 31 Dec 2026 | $0.75 | $3.75 | $33.75 |
| Gemini gemini-3.8-flash, from 1 Jan 2027 | $1.50 | $7.50 | $67.50 |
| OpenAI gpt-6.1-terra | $1.00 | $5.00 | $45.00 |
| Claude Sonnet 5.5 | $2.00 | $10.00 | $90.00 |
| OpenAI gpt-6.1-sol | $2.00 | $10.00 | $90.00 |
| Claude Opus 5.5 | $4.00 | $20.00 | $180.00 |
Standard list prices, before caching, batch discounts and taxes. Not a measure of quality. Check each provider’s page for the current figures.
What it means per customer
At $90 for 10,000 requests, each request costs $0.009. Suppose a customer makes 200 requests a month: that’s $1.80. On an example $29 plan, that’s fine. A customer making 5,000 requests costs $45, which is more than they pay. That’s why you want limits in place from the start.
Four ways to lower the LLM API cost per month
- Use the cheapest model that passes your testsThe price gap between tiers is large, so test the cheap one first.
- Cache repeated textProviders charge much less for repeated prompt text. Long system prompts benefit most.
- Batch work that can waitAnthropic gives 50% off through its Batch API, and OpenAI’s batch rates are about half on most models.
- Cap output lengthOutput tokens cost several times more than input, so shorter answers save the most.
Common questions
A chunk of text that models count and bill by. A token is often part of a word. Providers offer tools to count tokens for your own text.
Tool use, web search, long context and failed retries can all add cost. Agents can make many calls for one user action, so measure real usage.
You should at least log tokens per customer from day one. Then you can see who costs what, and set limits based on facts. See [how to integrate AI into SaaS](/ai-mvp-development/add-ai-to-saas).
Keep reading
Tell us what you’re building.
Takes under a minute. We come back with a scope and a fixed price.
- Reply within 24 hoursA real person reads every brief.
- Fixed scope, fixed priceYou get a written proposal, no hourly billing.
- You own the codeFull ownership of code and assets at launch.