Chinese LLM API pricing comparison (2026)
Rough per-million-token figures for the four mainstream China models, and what a compatible gateway typically charges on top. Prices shift often — treat this as directional.
The raw model costs
| Model family | Input / 1M tok | Output / 1M tok | Note |
|---|---|---|---|
| Doubao (Lite) | ~¥0.6 | ~¥3.6 | Cheap general workhorse |
| Qwen (Turbo) | ~¥0.3 | ~¥0.6 | Best price/perf |
| GLM (Air) | ~¥0.6 | ~¥0.6 | GLM-4-Flash free tier exists |
| Hunyuan (TurboS) | ~¥0.8 | ~¥2.0 | Strong reasoning |
At USD conversion that's roughly $0.04–$0.30 per million output tokens — a fraction of typical US-model pricing at similar quality.
What a gateway charges
A compatible gateway (like TideLink) adds a margin on top for the access, normalization, failover, and support. The point isn't that the gateway is cheaper than raw — it's that it removes the account, payment, and integration cost that would otherwise block you entirely.
Free models change the math
Some China models ship free tiers (e.g., GLM-4-Flash, Hunyuan-lite). A gateway can route low-stakes traffic there at near-zero cost, reserving paid models for harder tasks — your effective blended price drops further.
TideLink · All guides · TideLink is operated by Yuncheng Yanhu Beicheng Chaoxi Network Technology Studio, a sole proprietorship registered in Yuncheng, China (Unified Social Credit Code 92140802MAKM59LT6K), providing software development and IT integration services. Not a resale of third-party credentials.
中国注册主体：运城市盐湖区北城街道潮汐网络科技工作室（个体工商户，统一社会信用代码 92140802MAKM59LT6K）
Top comments (0)