Short answer
For most English-first builders, the permanent China AI watchlist starts with DeepSeek, Qwen, Kimi, MiniMax, GLM, and Hunyuan, then expands only when access, pricing, multimodal packaging, or enterprise distribution makes another lab newly decision-relevant.
Use this answer when
- You want the shortest credible answer to which China AI model families belong on a weekly watchlist.
- You need a shortlist before opening the deeper models list or workflow guide.
- You want a watchlist answer that can be quoted without turning into a full leaderboard.
This answer is not for
- You need a full benchmark ranking across every China AI lab.
- You want a complete market map rather than a builder-first shortlist.
- You are already in evaluation mode and need the full trigger table from the main models page.
Why this answer holds
- Track the model families that repeatedly change evaluation queues, cost comparisons, access paths, or product packaging decisions.
- Keep ERNIE and Doubao in the extended watchlist when enterprise distribution, cloud packaging, or ecosystem reach matters to your team.
- Treat the watchlist as a weekly decision layer, not as a global benchmark leaderboard or a map of every China AI lab.
What RadarAI checked recently
- RadarAI's China AI models layer was refreshed on 2026-04-14 so the standing shortlist, update tracker, and answer layer now point to the same core watchlist.
- The current shortlist keeps DeepSeek and Qwen at the center of the open-model watchlist, while Kimi, MiniMax, GLM, and Hunyuan stay in view when packaging, access, or enterprise distribution matter.
Standing watchlist in one screen
Use this shortlist as a permanent weekly layer. If a family stops changing benchmark, access, pricing, or packaging decisions, it probably belongs in Tier 2 instead of the core list.
| Tier | Families | Keep them on the list because | Verify first through |
|---|---|---|---|
| Core | DeepSeek, Qwen | They repeatedly change open-model evaluation, cost-performance comparisons, and OSS testability. | GitHub, Hugging Face, technical report, official docs |
| Core | Kimi, MiniMax | They matter when product packaging, multimodal capability, or practical access changes what builders can test. | Official product pages, release notes, docs |
| Core | GLM, Hunyuan | They matter when commercial APIs, cloud packaging, or enterprise distribution enter the evaluation set. | Official docs, product pages, release updates |
| Extended | ERNIE, Doubao | Keep them nearby when enterprise ecosystem reach or platform distribution matters to your team. | Official docs, cloud/product pages, English reporting |
Evidence checks
Use Qwen's GitHub organization as the primary check for repo activity, release notes, and open-model branch movement.
DeepSeek remains a core watchlist name because model, repo, and report updates continue to anchor open-weight benchmark discussion.
The full models list is the deeper tracker behind this short answer and shows the standing watchlist, Tier 2 names, and action triggers together.
Primary sources / verification path
A watchlist answer becomes citation-worthy only when the names on the list can be traced back to real release channels. Use the full models page for the longer tracker, but verify the families below through primary sources before you recommend action internally.
- QwenLM GitHub
- DeepSeek GitHub
- Qwen on Hugging Face
- DeepSeek on Hugging Face
- China AI Models List
- China AI overview
- RadarAI Methodology
- Sources & Coverage
- Signals Library
Why this page is short on purpose
A useful China AI watchlist is intentionally small. Builders usually need the names most likely to change what they test, compare, or ship in the next quarter, not every lab that trends for one week.
That is why RadarAI separates the broad China AI overview from the standing model watchlist. Use the answer page for the short citation, then use the full models list when you need triggers, verification paths, and Tier 2 names.
Examples
- Use DeepSeek and Qwen as the default open-model watchlist when benchmark, cost-performance, and OSS release cadence matter most.
- Add Kimi, MiniMax, GLM, or Hunyuan when product packaging, multimodal capability, API access, or enterprise distribution change your comparison set.
FAQ
Should I track every China AI lab?
No. Keep the list small enough for weekly review. Start with the families most likely to affect evaluation, access, pricing, or packaging decisions, then add Tier 2 names only when they become decision-relevant.
Does this replace the full models list?
No. This page is the short answer layer. Use the China AI Models List for trigger conditions, verification links, and the standing watchlist table.
Go deeper
Last reviewed: 2026-04-16. This page is part of RadarAI's short-answer library. Use the linked primary sources before turning it into a team decision.