Just as the choice of operating system affects your budget and control, the type of AI models you use can shape your ability to experiment, innovate, and succeed, especially if you’re an early-stage entrepreneur or student.
This section walks through key considerations, practical limitations, and scenarios where open-source AI shines, helping you decide what gives you the most agency and capability.
Your first question is budget. What can you afford to spend, and what are the alternatives?
Free proprietary models:
- OpenAI ChatGPT, Google Gemini, Anthropic Claude, Perplexity, and Grok all have free versions.
- You may be able to stack multiple of these and use the free tier access to these models for free.
- Things to be aware of: You cannot toggle on and off the security settings on these models, but if you’re only working with public data, it doesn’t necessarily matter.
- Another option is to bounce between models. Use the power of each model until it runs out (as they are usually limited), then switch to the next model.
Paid proprietary models:
- Like any sport, you’re probably best off prioritizing some funds for a professional AI model; the model you pick is up to you. We’ll cover that separately, but you should consider one or maybe two models.
- Other hacks you can consider:
- Tools like https://aistudio.google.com/app/u/1/prompts/new_chat give you system prompt access to a range of tools that are free to use in test mode. This gives you full access to powerful model prompts essentially for free.
- API access to Google models, I believe, allows you to run 5 API calls a minute for free for testing and play. This is something you could leverage.
- You could build your own AI chatbots using third-party hosting tools (even free ones) with vibe coding, and only pay for the tokens you use. This would be a really cost-effective way of doing it.
- And more workarounds:
- Meta Llama runs inside the new Meta AI app and on Facebook. Llama is a pretty powerful model that is open source and free.
- You could consider using Mistral, which is a European model.
- You could use the free version of NotebookLM.
- You could use WolframAlpha for mathematics.
- You could also access Hermes, which is a free model based on Llama.
- There are also Chinese models like Qwen from Alibaba and Deepseek.
You’ll need to be cautious and hyper-aware of the terms and conditions for all these models and how you’re accessing them. You could also use more advanced AI models like Trae for queries/prompts. Access to core models is pretty liberal these days, and you can always use models until you hit limits and move on to the next one.
Home hosting of models
Yes, you can go to Hugging Face and download models to run on a local machine, and theoretically, these models are free to use. The downside of this approach is that you will probably need powerful computational resources or use a very small model optimized for low-end hardware. It’s not impossible, however, particularly if you have access to low-end hardware.
There are a few things to track here:
- While $20 a month for an AI model doesn’t seem like much, if you start stacking them, the expenses quickly add up.
- Think strategically about what you’ll use and what you don’t.
- Track your own use and figure out if paying for a model is really worth it.
- For instance, Perplexity gives you 5 pro searches a day. On that platform, that might be enough to meet your needs, and the non-pro searches are still good. So don’t instantly turn away if something is paid.
- Personally, I’d look to implement systems thinking when it comes to AI. What’s your stack, and what do you want to master?
- E.g., you might want a tool like RooTools or Trae as a code builder, or Lutra as an AI agent. You probably want a vibe coding tool like Replit.com. And you might want Anthropic Claude and a Grok account tied to a pro X.com account—either standalone or part of Grok.
- You might want to attach a Google Gemini account to a Google One account and have that handle your productivity suite as well, and dump Microsoft Office. That’s what I’d do.
- Canva is also an awesome tool to consider.
- You want to carve out your own little toolbox and drive forward that way.
Conclusion
The ultimate smart move is to look at your technology and lifestyle as a package
E.g., if you know AI is going to accelerate you forward, generate revenue, and grow your potential faster, a nominal investment may be negligible compared to what it gets you. It could mean an entirely new career path for you.
Compare it to the cost of self-education or going to university. On that basis, it’s cheap. It’s also entertainment and provides other utility.
I’d look to minimise the amount you spend on hardware. E.g., switching from an iPhone to a mid-spec Android phone would up your AI game in one move. Use the savings and spend it on the tooling.
Go open source for your general software to save money there and reallocate it to AI investment. Pair it with a Linux or ChromeOS machine, and you’ll stretch your dollar further.
And dare I say it, perhaps cut out the computer game costs and allocate them to AI. Shift your focus to something real.
I wouldn’t spend too much time messing around with open-source pre-trained models on Hugging Face Transformers, etc., only because the compute cost to run them will probably offset just going to the cloud as a user.