Open source vs paid AI tools and what their licences mean
Two questions decide everything: what does the licence let you do with the tool, and what do you pay once the free tier ends? "Open source" and "free" answer neither of them properly. This page does.
All the figures below are the ones we checked on 18 October 2026 and published in Top Free AI Tools and AI Pricing Matrix; the star counts come from the GitHub API.
What the common licences actually mean
| Licence | What you can do | What you must do |
|---|---|---|
| MIT | Use it, change it, sell it, ship it closed. | Keep the copyright notice and the licence text. |
| Apache-2.0 | The same as MIT, with a clearer patent grant. | Keep the notice, state what you changed. |
| GPL-3.0 | Use it and change it; you can charge for it. | If you distribute your modified version, its source must be under the same licence. |
| AGPL-3.0 | The same, including running it as a service. | If users interact with it over a network, you have to offer them the source. |
| Source-available | Usually read it and run it for yourself. | Nothing is guaranteed: the terms decide. Often no reselling, no hosting as a competing service. |
Two practical consequences. First, if you are a business, MIT and Apache-2.0 are the easy ones; GPL and AGPL matter the moment you ship or host a modified version, not before. Second, "source-available" is not open source, however open the repository looks. A README that says "open" is not a licence.
Real examples, with licences and star counts
This is the part that surprises people. Of the local-AI projects we list, MIT and AGPL sit side by side, and two of the most popular ones are not open source at all.
| Project | Licence | Stars | What it is |
|---|---|---|---|
| Ollama | MIT | 183.7k | Run open-weight models locally with one command. |
| Open WebUI | Source-available | 156k | Self-hosted chat interface for local and API models. |
| ComfyUI | GPL-3.0 | 138.1k | Node-based image and video generation. |
| llama.cpp | MIT | 131.5k | LLM inference in C/C++, CPU or GPU, no Python stack. |
| vLLM | Apache-2.0 | 94k | Serving engine for many requests per second. |
| Text generation web UI | AGPL-3.0 | 47.7k | Desktop app for local models, with an OpenAI-compatible API. |
Star counts are popularity, not quality: they measure how many people clicked a button, not whether the tool fits your machine or your licence rules. They are still the cheapest signal for "is anyone maintaining this", which is why we publish them with the date we read them instead of an opinion.
What the paid side really costs
On the paid side the trap is the opposite one: the headline is "free", and the useful version is not. These are the published prices we could verify, in US dollars, per month:
| Service | Free plan | First paid tier | Top tier |
|---|---|---|---|
| ChatGPT | Yes | $20 (Plus) | $200 (Pro) |
| Claude | Yes | $17 (Pro, billed annually) | Max, from $100 |
| GitHub Copilot | Yes, limited chat and agent use | $10 (Pro) | Pro+ $39, Max $100 |
| Cursor | Yes | $8 (Plus) | Premium $30, Super $100, Ultra $200 |
| ElevenLabs | Yes | $6 (Starter) | Creator $22, Pro $99, Scale $299, Business $990 |
Read the Claude row twice, because it is the most common source of confusion: $17 is the monthly price when billed annually. Paid monthly it is $20. The same trick appears everywhere — annual billing is cheaper per month, and taxes or local currency are not in these numbers.
How to decide in two minutes
- Do you need to modify it, ship it or host it? If yes, read the licence first, and remember AGPL's network clause.
- Do you need it on your machine, offline? Then check the hardware before the licence: our VRAM calculator says whether it fits.
- Do you only need answers? Then the free plans and free APIs are almost always the cheapest route, and the licence is irrelevant.
- Is the free plan enough? Test it for a week before paying. Limits are what the paid tier removes, and they change often.
Frequently asked questions
Is open source always cheaper than paid?
No. The licence costs nothing, but your time and your hardware do. Running a local model needs a graphics card and maintenance; a $20 subscription needs neither.
Can I use open-source AI tools in a company?
With MIT and Apache-2.0, yes, almost without thinking. With GPL and AGPL, the licence applies to what you distribute or host, so check before putting it in a product. With source-available projects, read the terms — that is precisely what the licence is restricting.
Why do you publish stars rather than a score?
Because a score would be our opinion, and a star count is a fact with a date attached. If a number changes, we change the number.
How current are these prices?
They were read from each provider's own pricing page on 18 October 2026. They change without notice, so confirm on the official page before you commit to a yearly plan.
Keep reading
- AI Top Picks — the tools we would actually recommend, by task, with licence and price.
- Top Free AI Tools — 28 open-source projects and 8 free plans, with stars and licences.
- AI Tool Comparison — two tools side by side, on the criteria that matter.
Sources and date: licence terms are the ones published in each project's repository; star counts come from the GitHub API; subscription prices come from each provider's official pricing page. All read on 18 October 2026. Nothing here is legal advice — for a real compliance decision, read your own project's licence file.