Blog
Notes on open models: what they are, how to try them, and what we learn running them.
tryingopen
Meet tryingopen, a free way to try open models without an account
Tryingopen.com is Offline Base's free chat for 24 open models, with no account. A first visit, from the default pick to sharing a chat as a link.
Offline Base
Our default model runs on a palm-sized Jetson Orin Nano
The board, the model and the path behind tryingopen's default pick, Liquid AI's LFM2.5-VL 3B on an NVIDIA Jetson Orin Nano, as of October 2026.
Offline Base
How the d1 router picks between our small model and the cloud
Liquid AI's d1-3B runs on one of our Jetsons and decides, message by message, whether our small model or a cloud model answers you.
Offline Base
Our 3B model's answers, set beside any cloud model you pick
Pick a cloud model and your first two questions also go to Liquid AI's 3B model on our edge devices, with both answers set side by side.
Models
A tour of the 24 models in the picker
How the tryingopen model picker is grouped, what the $ bands and badges mean, and how to pick a model for pictures, files or long documents.
Models
What's new in the picker since September
Six of the models added to the tryingopen picker in September and October 2026: what each maker released, and how each one runs on the site as of October 14.
Models
LFM2.5 24B joined the picker the day Liquid AI released it
Liquid AI released LFM2.5-24B-A2B on October 13, 2026, and it joined our picker that day. What Liquid says about it, and how we fixed its search.
Models
Gemma 4 on tryingopen, from our boards to the cloud
Google's Gemma 4 comes in three sizes on tryingopen: E2B on our own Jetson boards, and the 26B and 31B in the cloud.
Guides
What total and active parameters tell you about a model
The size in our model picker is mostly a total. What total and active parameters mean for memory, for work per token and for the sizes other sources quote.
Guides
What counts as an open model, and what the licenses allow
OSI's definition of open source AI, why the models on tryingopen are open-weight, and what each license lets you do with the weights.
Privacy
What zero data retention covers, and where it stops
Zero data retention is a host's promise to keep none of your messages. How tryingopen enforces it on each request, and the five parts of the site outside it.
tryingopen
Choosing the company that runs your model
Since September 20, 2026, you can pick which company runs a model on tryingopen: how the list is built, what a pick does and what "Provider may log" means.
Privacy
The four things tryingopen keeps about your visit
A plain reading of the tryingopen privacy policy and terms, covering the few things Offline Base keeps and what stays in your browser or on your device.
Privacy
Sharing a chat, a model setup or a preset as a link
A share link holds a whole chat after the # in the address, a setup link names your models, and presets stay in your browser. Here is what reaches our servers.
Privacy
Self Chat saves your chats only where you choose
Self Chat keeps each chat in your Google Drive or a folder on your computer, and sends each cloud message under a zero-retention requirement.
Guides
Use Ollama, LM Studio, llama.cpp or Jan from Self Chat
Connect storage, start a model server, let the site in, and chat with a model on your own computer from tryingopen's Self Chat.
tryingopen
How web search works on tryingopen
The three ways a reply on tryingopen can search the web, why each one is capped, where search stays off, and how it sits outside zero retention.
tryingopen
Sending pictures and PDFs, and getting files back
What you can attach on tryingopen, which 13 models read pictures, how PDFs are read, and how files a model makes open safely in your browser.
tryingopen
Ask a closed model the same question with Compare
Under any finished reply, Compare asks Claude Sonnet 5 or GPT-5.6 Terra the same question, so you can judge an open model on your own task.
tryingopen
Reading a reply: its model, provider, speed, cost and code
Each tryingopen reply names its model and, most of the time, who served it, its timings, tokens, cost and the request to run it yourself.