Nine Chinese AI models and tools.
Which one fits your work?

The nine names in this collection do not describe nine identical kinds of product. Some are individual models that developers can call through an API. Others are applications that package several models, search, file handling and creative tools into an interface. That distinction matters more than a large parameter count or a “free” badge.

A practical comparison asks three questions: can it handle your input, can it produce the output you need, and what does a complete task cost after revisions? The profiles in this directory answer those questions where the providers document them, and identify what remains unverified.

Start with the task

For software work, explore MiMo-V2.6-Pro, GLM-5.3-Flash and DeepSeek-V4.1-Flash. Their documentation describes visual understanding alongside coding or agent workflows. This makes them candidates for comparing an implementation with a screenshot or working from mixed inputs; it does not prove which will perform best on your project. Xiaomi release ↗ · GLM documentation ↗ · DeepSeek release ↗

Ling 3.1 Flash is another candidate for text-heavy coding and analysis. For research that should become a report, presentation or spreadsheet, Kimi offers dedicated agent and office workflows. Review the generated references and calculations before treating a polished file as finished. Ling provider details ↗ · Kimi agent overview ↗

For images and short clips, Dreamina is the most directly creative product in this selection. Qwen Studio combines broader assistant and creative modes. ERNIE / Wenxin and Doubao are also worth examining for assistant tasks and Chinese-language material. The exact model, interface and enabled features matter in each case; the profiles link to their official documentation.

“Free” can mean several different things

Dreamina’s free credits are an allowance. Kimi’s advanced features draw on plan credits. A promotional API route may stop working or start billing when its offer ends. A free chat account and a developer API can have entirely separate terms. Released model weights are another category again: running them requires hardware and hosting. Dreamina plans ↗ · Kimi credit rules ↗

Before choosing a tool, check the allowance for the exact feature you want. Count the attempts needed to get a usable result, not just the cost of one generation. For an automated task, set a usage budget and confirm what happens when the free allowance runs out.

Understanding an image is different from generating one

A multimodal model can read a screenshot yet return only text. GLM-5.3-Flash explicitly documents that combination. An agent may use its text or code output to call an image tool, render a page or build a presentation. Check the whole workflow before assuming that a model itself generates every file shown in a demo. Input and output modalities ↗

Run a small comparison before committing

Choose one task you understand well. Give two candidates the same inputs, constraints and success criteria. For code, run the program and tests. For research, verify the sources. For visuals, inspect text, object details and continuity. Note time, total usage and corrections needed. Repeat with a second example so one lucky result does not decide the comparison.

The best choice is the one that reliably completes your work within your budget and review process. The directory offers a starting point, with separate profiles for all nine products, practical strengths and limitations, and dated access notes.

This is a documentation-based guide. We have not run a common benchmark across these products. Availability and pricing can change after the review date.

Explore all nine profiles ↗

View the supplied collection cover

Reference material retained unchanged. Its blanket free-access claim is not the conclusion of this guide.