Home / Chinese AI models

Chinese AI models

Eight Chinese labs now sell frontier-class models to developers worldwide, most at a fraction of US prices. Here’s every one we track, checked November 17, 2026.

Prices checked

Quick answer

Qwen4 Max (Alibaba) is the newest Chinese flagship, at $1.50 / $4.50 per million tokens. Kimi K3.1 Pro and GLM-5.3 target coding and agents, DeepSeek V4 and MiniMax M3 are popular open-weight models, and Tencent Hy3 is the cheapest at $0.132 / $0.528.

Every Chinese model we track

International (non-mainland China) API prices, per million tokens.

Chinese AI model prices
ModelContextInput $/1MOutput $/1MBest for
Kimi K3.1 Pro
Moonshot, China
Open weights
1M$2.00$10.00Coding and agents; weights released Nov 5
Qwen4 Max
Alibaba, China
New
1M$1.50$4.50Alibaba’s newest flagship, now on pay-as-you-go
GLM-5.3
Zhipu, China
Open weights
1M$1.40$4.40Zhipu’s newest flagship, strong at coding
DeepSeek V4 Pro
DeepSeek, China
Open weights
1M$1.32$3.96Strong open model; half price off-peak ($0.66 / $1.98)
Doubao Seed 2.0 Pro
ByteDance, China
256K$0.50$3.00Model behind China’s most-used chatbot, Doubao
DeepSeek V4.1 Flash
DeepSeek, China
Open weights
1M$0.30$1.20Fast open model; half price off-peak ($0.15 / $0.60)
MiniMax M3
MiniMax, China
Open weights
1M$0.30$1.20Low-cost model that reads images and video; community license
MiMo-v2.6 Pro
Xiaomi, China
Not stated$0.435$0.87Very low-cost reasoning model from Xiaomi
Hy3
Tencent, China
Open weights
256K$0.132$0.528Cheapest open-weight model here (Apache-2.0)

Questions people ask

What is the best Chinese AI model?

It depends on the job. Alibaba’s Qwen4 Max is the newest flagship, Kimi K3.1 Pro and GLM-5.3 are strong at coding, and DeepSeek V4 Pro is the best-known open model.

Are Chinese AI models open source?

Many are. DeepSeek V4, Kimi K3.1 Pro, GLM-5.3, MiniMax M3 and Tencent Hy3 publish their weights, though licences differ. Qwen4 Max and Doubao Seed are API-only.

Are Chinese AI models cheaper?

Usually. Most cost between $0.50 and $10 per million output tokens, compared with $10 to $50 for the top US models.