← All guides

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 familyInput / 1M tokOutput / 1M tokNote
Doubao (Lite)~¥0.6~¥3.6Cheap general workhorse
Qwen (Turbo)~¥0.3~¥0.6Best price/perf
GLM (Air)~¥0.6~¥0.6GLM-4-Flash free tier exists
Hunyuan (TurboS)~¥0.8~¥2.0Strong 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.

See live pricing

Get an API key →