How to reduce API costs and AI token usage in CheapAI
2 minute read
AI models view everything as a sequence of tokens. For example, they perceive the sentence 'Croatia has beautiful beaches' as 'Croat|ia|has|beaut|i|ful|beach|es'.
The cost of one response you receive is: the cost of reading input tokens + the cost of generating output tokens.
Before responding to your message, AI models read the entire chat from scratch. Every time. This costs! The longer the chat, the more expensive new responses become.
Prefer Short Conversations ✂️
Start a new conversation instead of extending the old one. If you must, use the tactics described below.
Reduce Memory 🧠
Often it's enough for the model to look at only the last few messages instead of the entire conversation. This allows you to have infinite conversations without increasing the cost.
Enable Auto Caching (with caution) 💾
Not all models can do this. If the time gap between your messages is small, this can reduce costs by 50% - 90%. If the gap is not small, it will increase costs! Additionally, the first message you send will always be slightly more expensive.
Edit Message Instead of Sending a New One ✏️
When you're not satisfied with the received response, edit your message and regenerate it. This will prevent the conversation from growing.
Maximize Use of Official apps 💻
Some models can be used for free in official applications. Make the most of them, and jump into CheapAI when you encounter limitations.
Monitor Message Costs 💸
Above each message you receive, CheapAI shows you how much it cost. Watch it!
Use English and Latin Alphabet 💬
English uses the least tokens and money. For example, Arabic is 3x more expensive (try it here). Tell the model to always respond in English, and try your best to do the same.
Set System Instructions ⚙️
For example, you can write 'respond briefly and directly' or 'respond with code only'.
Ask Multiple Questions in the Same Message 🔢
Don't send multiple messages in a row with one question.
Use Free Models 🎁
Many models are free to use in CheapAI, check within each provider. The most powerful Gemini models by GoogleAI are free in certain countries.
Sometimes Use Cheaper Models 👶
Consider whether you really need the strongest model for a particular task. In CheapAI, you can change the model mid-conversation.
Create a Conversation Summary 📝
Tell a cheap/free model to create a summary of the conversation. Then copy that into a new conversation and continue there.
Turn off Unnecessary Tools 🛠️
Even if the model does not use a certain tool, the message price will be higher just because it is enabled.