Open-weight Chinese models are still not the same as permissive open source
tamla_htoo · reddit · 2026-07-23
The post argues that many large Chinese open-weight models are being marketed as if they were open source, but in practice they are closer to closed models with a download option.
Main points:
- the models are often so large that only companies with serious GPU clusters can run them at scale
- many licenses allow commercial use, but restrict offering the model itself as a hosted inference service without permission
- that means the original model creator can still control who becomes a direct competitor
The author says open weights are still useful for research, fine-tuning, and building apps, but the industry should be precise: downloadable weights are not the same as permissive open source like Apache 2.0 or MIT.
Related event: Open Weights Are Not True Open Source in AI(2 posts)→
More from Models
- Infosec debate says next-gen AI incident response still lacks trust and audit logs — npinto · 2026-07-23
- OpenAI’s Codex is reportedly getting a real-time voice mode and personal assistant layer — tinyfool · 2026-07-23
- Kimi K3’s open-source release is drawing more excitement than Opus 5 — Angaisb_ · 2026-07-23
- A small Google version-number joke lands on Pro 3.5 and Flash 3.6 — TheZvi · 2026-07-23
- Claude Opus 5 rumored soon as Opus 4.8 starts routing to it — koltregaskes · 2026-07-23
- LLaDA2.2-flash brings 128K context and Levenshtein Editing to agentic diffusion models — pmttyji · 2026-07-23