Meet tryingopen, a free way to try open models without an account

Tryingopen is a free chat for open AI models. We build and run it at Offline Base. It has been live since August 17, 2026. You open the page and start typing. There is no sign-up form and no login.

The site's own description puts it in one line: "No account, nothing saved: pick a model, start chatting, and share any chat as a link."

The header and its two modes

Every page carries the same header: "Tryingopen: try any open model", with "by Offline Base" under it. The header also has two mode links.

• Try open is the home page at tryingopen.com. Its headline is "Try any open model." This is the ordinary chat.
• Self Chat lives at tryingopen.com/self-chat. It saves each chat only in a place you connect (Google Drive or a folder on your computer, as of October 2026), and it sends each cloud message with a zero-data-retention requirement attached.

The picker: 24 models from 13 labs

As of October 14, 2026 you can chat with 24 models in the picker, from 13 labs: Z.ai, Alibaba, NVIDIA, DeepSeek, Google, OpenAI, Meta, Thinking Machines, Tencent, Mistral AI, Xiaomi, inclusionAI and Liquid AI. The picker also lists two more, MiniMax M3 and GLM 5.2, greyed out with an "Adding capacity" badge.

Both models under Offline Base come from other labs, and we run them on our own hardware: Liquid AI's LFM2.5-VL 3B and Google's Gemma 4 E2B.

A note in the picker says how open the list is: "23 of these 24 are open: anyone can download and run them. The other is announced as open, with the weights still to come." That is why the site's description begins "Try 23 open AI models" while you can chat with 24. One of the 23, Qwen3.8 Flash, is Alibaba's hosted production model, built on the open-weight Qwen3.8-Flash-Next. What counts as an open model covers the one exception and each license.

The picker groups models under "Start here", "Quick and light" and "Big and powerful". Each row has a "Learn more" panel with what the model is, how its provider treats your messages and a link to the maker's announcement. A $, $$ or $$$ band shows roughly what a model costs us to run, next to the words "Free for you here." A tour of the 24 models in the picker explains the groups, the bands and the badges.

A first visit

When you arrive, a model is already chosen: LFM2.5-VL 3B, Liquid AI's small vision-language model. The line under the message box reads "Open model · Runs on an edge device". Hover over "Runs on an edge device" and the page explains it:

Quick questions are answered by a small model on Offline Base's own hardware; harder ones go to the cloud model you pick. Every reply says which one answered.

Every message you send goes first to d1, a small decision model from Liquid AI that we run on one of our own devices. Since October 14, 2026, d1 asks whether a message is simple enough for the small model. Quick questions, pictures and quick web lookups are answered right there by LFM2.5-VL 3B on our hardware. For a lookup, the search itself goes through OpenRouter to a search provider, outside the zero-retention requirement. Anything more than a quick question, such as code, maths, advice or a longer, detailed answer, goes to a cloud model. That cloud model is GLM 5.3 Flash from Z.ai unless you pick another in the "Cloud fallback" picker.

Each reply from this default pick draws its route: from you to the d1 router, then on to the edge device or the cloud. Its "Learn more" link opens "How the small model answers", where bars show d1's estimates for that message. The rest of what a reply shows is in What each reply tells you about how it was made.

You can switch to any of the other models from the picker. When you pick a cloud model, your first two questions in that chat also go to our small model, and you can open its answers beside your pick's (how the side-by-side works).

When a chat is worth keeping, the Share button puts the conversation's text into a link, leaving out pictures, files and reasoning. The chat rides after the "#", which browsers never send to a server, so we hold no copy, and anyone you give the link to can read it. A very long chat is too big for a link; export it as Markdown instead.

What the privacy policy says

Our privacy policy took effect on October 12, 2026. The short version:

• The chat sets no cookies.
• We store no conversations on our servers.
• We don't sell what you send, use it for advertising or use it to train AI models.

We keep four things: hourly message counts per IP address, which expire on their own within about two hours; page-view analytics that use no cookies; standard hosting logs with no message text; and, on our edge devices, the size and timing of each request (details on each).

Outside companies serve the cloud models, and the picker shows how each one handles what you send. Some models carry a "Provider may log" label, and GLM 5.3 Flash, the default cloud model, is one of them. If you pick a provider for it, your message goes with a zero-retention requirement instead (more on choosing a provider).

Posts on the small model, sharing, Self Chat and Compare

• The small model: Our default model runs on a palm-sized Jetson Orin Nano and How the d1 router picks between our small model and the cloud.
• Sharing: Sharing a chat, a model setup or a preset as a link.
• Self Chat: Self Chat saves your chats only where you choose.
• Compare: Ask a closed model the same question with Compare.