Models / Use cases
Free AI models, every way to run models at $0
There are three real ways to use AI models free: starter credits (every CompanyFabric account includes them, no card required), models whose providers run generous free tiers, and open-weight models you can download and run on your own hardware. These picks cover all three, ranked by how much free capability you actually get.
Ranked
The picks
GLM-5.3-Flash
Z.ai
Starter credit stretches furthest here, it is the cheapest model in the library by a wide margin.
Context
1M tokens
Max output
131K tokens
Tools
Yes
Vision
Yes
View GLM-5.3-Flash pricing and playground →
DeepSeek V4.1 Flash
The V4 family's volume tier, starter credits cover thousands of calls at its prices.
Replaces DeepSeek V4 Flash, retired by DeepSeek on 2026-09-10.
Gemini 3.1 Flash-Lite
Google's cheapest tier, and their free tier on this model is generous if you go direct.
Kimi K2.6
The agentic value tier, free starter credits stretch furthest on models priced like this.
GLM-5.2
The most generous free access in the frontier-adjacent class, and open weights if you outgrow the tier.
Retired
Previously recommended here
The V4 family's volume tier, starter credits cover thousands of calls at its prices.
Use DeepSeek V4.1 Flash instead, same endpoint, same key, one line changes.
FAQ
Frequently asked questions
Is there a catch to the free starter credits?
No card and no subscription. Welcome credit runs through companyfabric/auto, which routes each request to the cheapest model that can serve it, so the credit goes as far as possible. Naming a model directly needs a top-up, and any amount unlocks the full catalog.
Why does the free credit only work with companyfabric/auto?
Because it is what makes free credit sustainable without a card wall in front of it. Auto is also the product, so the free tier demonstrates the thing we actually sell, and you can see which model served every request in your usage.
Are open-weight models really free?
The weights are free to download and run; the GPU is not. Self-hosting a 100B-class model needs serious hardware, which is why most teams prototype through a metered API first and only move to their own GPUs at sustained scale.
What is the cheapest way to keep using a model after the free credit?
For steady usage, prepaid credits at metered per-token prices. If you already have a provider contract, BYOK routes on your key at 0% platform fee. Free credit, prepaid credit and BYOK all work behind the same API key.
Do the free credits expire?
Purchased credit never expires. Plan allowances run for their billing period, and your balance shows which is which, so you always know what you are spending.
One key. Every model. Exact prices.
Route every pick on this page through one key. Free starter credits included, no subscription required.