Open-weight AI models you can chat with free.

Open-weight models are AI models whose trained weights are published, so anyone can download, inspect or host them. UsingOpen is the hosted shortcut: it sends your chat to a live copy of each model, and shows you the size, context window and price before you start. Nothing to install, no account required.

The catalog currently has 23 models from 12 labs. At one end are small, sparse text models that answer almost instantly; at the other are trillion-parameter flagships built for long coding runs. Every model below is free to try when it is live; a model paused for capacity returns as soon as the provider brings it back.

Not sure where to start? GLM 5.3 Flash is the default and reads pictures and video; DeepSeek V4 Flash is the cheapest text-and-code option; Kimi K3 is the largest and most expensive.

All 23 open-weight models

How to read the table

• Parameters is the total size of the model. Active per token (shown on each model page) is how much of it runs for each word; a sparse model can have a huge total and still answer quickly.
• Context window is how much text fits in one conversation. A million tokens is roughly 750,000 words — a full codebase or a stack of long documents.
• Price / MTok is the upstream API rate per million tokens. On UsingOpen, chatting is free inside the daily prompt limit; this column is a fair way to compare what each model costs to run.
• Images means the model accepts pictures (and often video). Tools means it can call functions or run agent workflows. ZDR means the provider offers zero data retention.

Which open model should you choose?

The right model depends on the job, not on the leaderboard. These are the practical starting points:

• Coding and long projects: DeepSeek V4 Flash, GLM 5.3 Flash or Kimi K3 when the task is at the top end.
• Pictures, screenshots and video: Qwen3.8 Flash, Gemma 4 31B or MiMo V2.6 Flash.
• Very long text: any model with a 1M context — DeepSeek V4 Flash and Qwen3.8 Flash are the cheapest ways in.
• Privacy-sensitive work: models marked ZDR, such as Gemma 4 31B or MiniMax M3.
• Writing and editing: Gemma 4 31B is tuned for prose and reads images as well.

What free means on UsingOpen

You can chat with every model here without an account, up to 10 free prompts per user per day. There is no card and no paid tier. Answers stream in, reasoning models can show their thinking, and chat history stays in your browser rather than on our servers. If you want a second opinion, Compare mode asks two models the same question side by side.

Frequently asked questions

What is an open-weight model?

An open-weight model is one whose trained parameters are published, so anyone can download, inspect, fine-tune or host it. Some are fully open source; others, like Gemma, ship under their maker's own licence. UsingOpen does not make you host anything: it connects your chat to a hosted copy and shows the model, maker, size and price up front.

Are these models free to use on UsingOpen?

Yes. Every model in the catalog can be chatted with for free, without an account, up to 10 prompts per user per day. A model paused for capacity is unavailable until it returns, which the picker shows live. The per-million-token prices in the table are what upstream API access costs elsewhere; they are not charged on this site.

Do I need an account to try an open model?

No. There is no sign-up, email or phone number. Pick a model, type a message, and the answer streams back. Chat history is stored in your browser, not on our servers.

Which open model should I try first?

GLM 5.3 Flash is the site default: it reads text, pictures and video, holds a million tokens and is cheap to run. DeepSeek V4 Flash is the value pick for text and code. Qwen3.8 Flash is a strong multimodal all-rounder, and Kimi K3 is the largest model available when a task needs maximum capacity.

What do parameters and active parameters mean?

Parameters are the weights a model has learned; a larger number usually means more capacity, but not automatically better answers. Many models here are mixture-of-experts designs, which activate only part of the network for each token. That active count is a better guide to speed and running cost than the headline parameter total.

Why is a model greyed out or paused?

Availability depends on upstream capacity. When a provider pauses a model, the picker shows it as "Adding capacity" and it cannot be selected until it returns. The rest of the catalog stays usable, and the live status on the chat page is always the source of truth.

What is zero data retention (ZDR)?

ZDR is the provider's commitment not to retain the content of your requests after serving them. It is shown per model because it varies: some models here offer ZDR and some do not. It describes the upstream provider's terms, so check the hint before sending sensitive material.

Compare models head to head · All models · Compare models · Free AI chat · FAQ