Chinese Open-Weight Models Just Took Over HN — Here's What No-Code Builders Should Actually Use

Qwen 3.8 hit 925 points on Hacker News this weekend. Kimi K3 got so popular Moonshot AI literally suspended new subscriptions because they ran out of GPUs. Chinese open-weight models aren't just competitive with Western frontier models anymore. They're setting the pace on cost per token. For no-code builders, this reshapes what you can afford to build.

Last Saturday, the top of Hacker News wasn't a new JavaScript framework. It was a tweet from Alibaba's Qwen team announcing Qwen 3.8: 2.4 trillion parameters, open weights promised soon, no published benchmarks, and somehow 925 points and 656 comments.

Which Chinese open-weight models actually matter right now?

Three families dominate the open-weight space, and they serve different needs.

Qwen (Alibaba): Qwen3-8B is the workhorse small model. It won 6 of 13 hard evaluations against models 4x its size. Qwen3-Coder is the specialist coding model, 480B parameters in an MoE architecture, available on OpenRouter at $0.22 per million input tokens.

Kimi (Moonshot AI): Kimi K3 is the current benchmark leader among Chinese models, scoring 81 on BenchLM's July rankings, ahead of Qwen3.7 Max at 72.8. It's a 2.8T parameter multimodal reasoning model. Pricing is $3/M input, $15/M output. Moonshot paused new consumer subscriptions because GPU demand exceeded capacity. A software product, sold out.

DeepSeek: DeepSeek V4 Pro is the budget champion. They made the 75% discount permanent: $0.435/M input, $0.87/M output. DeepSeek V4 Flash goes even lower at $0.14/$0.28. For comparison, GPT-5.5 is about 12x more expensive at comparable intelligence levels.

What does this mean for no-code builders?

If you're building a no-code app and you need an AI feature, you now have three tiers of Chinese models to choose from, each cheaper than anything comparable from OpenAI or Anthropic. Budget tier: DeepSeek V4 Flash at $0.14/M input. Mid tier: Qwen3-Coder at $0.22/M input. Quality tier: Kimi K3 at $3/$15.

The takeaway

Chinese open-weight models are no longer the budget alternative. They're the default for cost-sensitive AI features, and they're closing the quality gap fast. DeepSeek V4 Pro at $0.435/M input versus GPT-5.5 at 12x the price is not a rounding error. It changes what kinds of AI features get built.

But the Kimi K3 subscription pause is the story I keep coming back to. A top-tier AI product literally couldn't accept new users because they ran out of GPUs. Use these models, save the money, but build with an exit plan.