# How to use top AI models on a budget ## Метаданные - **Канал:** Data School - **YouTube:** https://www.youtube.com/watch?v=wvvTog-FXOk ## Содержание ### [0:00](https://www.youtube.com/watch?v=wvvTog-FXOk) Segment 1 (00:00 - 05:00) If you're using the free versions of ChatGPT, Claude, or Gemini, you're not going to get the best results. So how can you get the best results without spending $20 or more per month? I'll show you. OpenAI, Claude, Google, and others all have better and worse AI models, and they limit your access to the best models if you're a free user. If you want to consistently use the best models, you do have to pay, but you don't have to buy their premium subscriptions. The secret is that there are actually two ways to pay for these models: monthly subscriptions and pay-per-token. Here's GPT-5, which is one of OpenAI's best models. What this tells us is that you can use this model if you pay them $1. 25 per million input tokens and $10 per million output tokens. Input is what you send to them and output is what they send back, and a token is 3/4 of a word on average. So a million tokens is a lot, I promise you. In order to pay per token, you need to have an API key. An API key is typically used by software programs to talk to AI models, but you can use API keys as well, and it's easy and free to get one. After you've signed up, you will need to fund your account with credits, and for OpenAI, the minimum is $5. Every time you use that API key to talk to a model, that credit balance goes down. But if you're a light to moderate user, $5 can easily last you months, and I'll show you some cost examples later. Anyway, let's say you've gotten your API key and you funded it with $5. How do you actually chat with GPT-5? That's where TypingMind comes in. TypingMind is an application where you enter your API keys from OpenAI, Claude, Google, and any other model provider, and it gives you the interface for chatting with the model. Just like ChatGPT, you've got past chats on the left, though I've organized mine into folders, and here's where you can chat with the model of your choice. I've only enabled the few models that I actually use, but you can enable any model you like. Now, there's one killer feature of TypingMind that you've got to try. But before I get to that, let's talk about cost, because you want to know how far your $5 will get you. Let me give you some examples. Here's a short chat about the grep command line tool that I had with GPT-5 Chat, which is a snapshot of ChatGPT. It cost about 1. 3 cents. In contrast, here's an incredibly long chat I had with GPT-5 over the course of about two months about how to format my upcoming Machine Learning book for print, PDF and ePub formats. It cost almost $4, but look how long this is, and we're not even halfway through the chat. Compare that to the $40 I would have spent on ChatGPT Plus for those two months in order to get such substantial access to this model, which is a reasoning model by the way. And now we're finally about halfway through the chat. Now let's talk about the killer feature of Typing Mind. You've probably been wondering which model you should use. Sometimes I have a preferred model for a given topic area based on my experience. For example, Claude is great for writing. But if you don't know which model you want to use and you really want a good answer, you can chat with multiple models at once. Here's my chat with GPT-5, Gemini 2. 5 Pro, and Claude Sonnet 4 about the subtitle for my Machine Learning book. I gave all of them info about my book and asked for subtitle suggestions and then down here were their responses. Then I picked my eight favorite ideas ### [5:00](https://www.youtube.com/watch?v=wvvTog-FXOk&t=300s) Segment 2 (05:00 - 07:00) and asked all of them to rank the ideas and tell me why they ranked them in a certain way. I found this invaluable for getting different perspectives on the same problem. Of course I'm paying the API costs for all three of these models, but it was only about 24 cents. And the beautiful part of TypingMind is that I get to choose which models I want to use and when and only pay for what I use. By the way, this is not a sponsored video for TypingMind. Finally, let's talk about the disadvantages of the approach I've shown here. Every model provider has certain features that you can only get if you use their UI and have a paid subscription. For example, Claude Pro includes Claude Code, their projects feature, their research feature integrations, and so on. The TypingMind team is aware of this and often releases their own version of these features. For example, TypingMind includes its own Deep Research tool. It has a knowledgebase feature so you can share your documents with any model, and so on. But if you strongly prefer the native features offered by Claude or OpenAI or Google, then my approach probably won't work for you. Also, TypingMind is not a free product, and personally I use the Premium version that I bought during a sale in order to access all of its advanced features. At this moment, Premium is on sale for $99, which is a one time payment for a lifetime license. That might feel like a lot, but for me it has paid off many times over by allowing me to access the top models without any recurring subscriptions. There are probably alternatives to TypingMind that are cheaper, but I've never looked because TypingMind has been such a great app, with new features being released all the time. if you're a TypingMind user or you use an alternative to TypingMind to access AI models, please let me know in the comments how your experience has been. And finally, if your ears perked up when you heard me talk about my upcoming Machine Learning book, I'd love for you to download the first three chapters for free and I'll let you know the moment I publish it. The link is in the description below. Thanks so much for watching and I'll see you in another video. --- *Источник: https://ekstraktznaniy.ru/video/23704*