🪙 Run your bots for less
Run your bots for less
Every time a bot runs, an AI service bills you, or eats into your plan's limits. These 8 guides show how to pay less for the same results, with prices taken only from the providers' official pages.
8 rules that cut your AI bill
- Use the smallest model that does the job well. How →
- Test on a free tier first, without private data. How →
- Check whether a flat monthly plan or pay-per-use is cheaper for you. How →
- Cache the parts of the prompt that never change, and batch what can wait. How →
- Run simple, private jobs on an open model on your own computer. How →
- Run bots only as often as things change, or only when you ask. How →
- Keep prompts, tools and pasted pages short. How →
- Set a spend limit before anything runs on autopilot. How →
The guides
-
🪙 Run your bots for less
Use a small model for simple jobs and save the big one for hard ones
Most bot jobs (sorting email, summarizing, checking a page) don't need the most powerful model. The small models cost a fraction as much, so route easy work to them and only send hard problems up.
Popularity: not measured -
🪙 Run your bots for less
Build and test your bot on a free tier before you pay
Several AI providers let you try their API or app for free. Use that to prove your bot works before you put a card on file, and know what you give up on the free tier.
Popularity: not measured -
🪙 Run your bots for less
Decide between a monthly AI plan and paying per use
A monthly plan (like Claude Pro) gives you a flat price with usage limits. Paying per use through an API costs exactly what you use. Which is cheaper depends on how much your bots run.
Popularity: not measured -
🪙 Run your bots for less
Cut your AI bill with prompt caching and batch discounts
If your bot sends the same long instructions every time, caching makes the repeat part much cheaper. If results can wait a few hours, batch processing halves the price at several providers.
Popularity: not measured -
🪙 Run your bots for less
Run open models on your own computer (or a cheap host) for simple jobs
Open-weight models can run on your own computer with a free tool like Ollama, so routine jobs cost nothing per message. Some hosts also run open models for a fraction of big-model prices.
Popularity: not measured -
🪙 Run your bots for less
Stop always-on bots from burning tokens: run them less often, or on demand
Every scheduled run costs tokens, even when nothing has changed. Match the schedule to how often the thing really changes, and switch fun-but-idle bots to run only when you ask.
-
🪙 Run your bots for less
Keep your bot's prompts, tools and context short
You pay for everything the model reads on every run: your instructions, tool definitions, pasted pages and the conversation so far. Trimming those is often the easiest saving.
Popularity: not measured -
🪙 Run your bots for less
Set spend limits and usage alerts on your AI accounts
Before a bot runs on autopilot, put a hard cap on what it can spend and check usage weekly, so a runaway loop can't run up a big bill.
Popularity: not measured
Cheapest model we list at each provider
Standard API price per 1 million tokens, from official pricing pages, checked Oct 21, 2026. Cheapest isn't always good enough: test it on your job first.
| Provider | Model | Input | Output |
|---|---|---|---|
| OpenAI | gpt-6-luna | $0.10 | $0.50 |
| Anthropic | Claude Haiku 5.5 | $0.10 | $0.50 |
| gemini-3.1-flash-lite | $0.25 | $1.50 | |
| xAI | grok-4.3 | $1.25 | $2.50 |
| Groq | openai/gpt-oss-20b | $0.075 | $0.30 |
| Mistral | Mistral Large | $0.50 | $1.50 |
| Meta | Muse Wonder | $4.00 | $20.00 |
| DeepSeek | deepseek-flash (V4.1-Flash-1017) | $0.30 | $1.20 |
See the full AI price table See what each blueprint costs to run
Questions about AI costs
How do I reduce AI API costs?
Use the smallest model that does the job well (small models can cost up to 100× less than the flagship from the same company), cache instructions that never change, send jobs that can wait to a batch API (50% off at Anthropic, Google, Mistral, OpenAI), keep prompts short, run bots only as often as things change, and set a spend limit before anything runs on autopilot.
What is the cheapest LLM API?
Of the models in our price table, the cheapest per token is Groq openai/gpt-oss-20b at $0.075 per 1M input tokens and $0.30 per 1M output tokens (official pricing page, checked Oct 21, 2026). Cheapest isn't always good enough: test it on your real job first.
How much does an AI agent cost to run?
It depends on how often it runs and how much it reads. Example: a morning inbox briefing that reads about 15,000 tokens and writes 800 on 22 weekdays a month costs about $0.03 a month on Groq openai/gpt-oss-20b, and about $4.18 on Anthropic Claude Fable 5.1, at standard API prices. Use the calculator on this page for your own numbers.
Is a monthly AI plan cheaper than paying per use?
If you run a few small bots, pay-per-use API prices are often a few dollars a month or less. App plans charge a flat monthly fee with usage limits, and include the app itself. Compare both before you commit: see our guide “Plan or pay-per-use: which is cheaper for you?” on this page.