Access Chinese AI Models via One API - Qwen, DeepSeek, Kimi, ByteDance Seed, MiniMax, GLM, StepFun, MiMo and Hunyuan

Current AIsa guide to Chinese LLM and media models: exact model IDs, context windows, capabilities, endpoints, and pricing links for Qwen, Wan, DeepSeek, Kimi, ByteDance Seed, Seedream, MiniMax, GLM, StepFun, Xiaomi MiMo, and Tencent Hunyuan. OpenAI-compatible through https://api.aisa.one/v1.

AIsa gives you one API key for China’s leading AI model families: Alibaba Qwen and Wan, DeepSeek, Moonshot Kimi, ByteDance Seed and Seedream, MiniMax, Zhipu GLM, StepFun, Xiaomi MiMo, and Tencent Hunyuan. Use the same AIsa wallet, usage logs, and OpenAI-compatible SDK setup you use for GPT and Claude.

This page was refreshed from the live AIsa Model Gateway catalog on September 30, 2026. For current prices, check aisa.one/models; the pricing guide explains how each model type is billed.

API base

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_AISA_API_KEY",
    base_url="https://api.aisa.one/v1"
)

response = client.chat.completions.create(
    model="qwen3.7-max",
    messages=[{"role": "user", "content": "Compare Qwen, DeepSeek, and Kimi for a coding agent."}]
)
print(response.choices[0].message.content)

Chinese text, vision, coding, audio, and video-understanding models use POST /v1/chat/completions, and most also accept Anthropic-compatible POST /v1/messages. POST /v1/responses is now served by most Chinese text-model routes (including qwen3-max, MiniMax-M3, deepseek-v3, glm-5.2, and kimi-k3) — exceptions include qwen-plus-2025-12-01, seed-2-0-mini-260215, and the step-3.5-flash routes; an unsupported combination returns 400 model_route_not_supported. Image models use POST /v1/images/generations and POST /v1/images/edits; Wan and Dreamina video models are asynchronous through POST /v1/video/generations. Use the exact model IDs below.

Alibaba Qwen and Wan

Model ID	Context	Capabilities	Best for
qwen-flash	1,000,000	Audio, Coding, Text, Video, Vision	Lowest-cost high-volume multimodal work
qwen-mt-flash	1,000,000	Text	Fast translation
qwen-mt-lite	1,000,000	Text	Lowest-cost translation
qwen-plus-2025-12-01	1,000,000	Audio, Coding, Text, Video, Vision	General multimodal workloads
qwen3-coder-480b-a35b-instruct	262,144	Coding, Text	Maximum Qwen coding capability
qwen3-coder-flash	1,000,000	Coding, Text	Fast coding pipelines
qwen3-coder-plus	1,000,000	Coding, Text	Balanced coding agents
qwen3-max	262,144	Audio, Coding, Text, Video, Vision	Strong general Qwen model
qwen3-vl-flash	131,072	Coding, Text, Video, Vision	Low-cost vision and video understanding
qwen3-vl-flash-2025-10-15	131,072	Coding, Text, Video, Vision	Pinned Qwen VL Flash version
qwen3-vl-plus	131,072	Coding, Text, Video, Vision	Stronger document/spatial vision
qwen3.6-plus	1,000,000	Coding, Text, Video, Vision	Long-context Chinese and bilingual work
qwen3.6-plus-2026-04-02	262,144	Coding, Text, Vision	Pinned Qwen 3.6 Plus version
qwen3.7-max	1,000,000	Coding, Text	Frontier Qwen reasoning and agentic coding
qwen3.7-max-2026-06-08	—	Text	Pinned Qwen 3.7 Max version
qwen3.7-plus	—	Text	Lower-cost Qwen 3.7 route
qwen3.7-flash	—	Text	Fast Qwen 3.7 route
qwen3.7-flash-2026-07-15	—	Text	Pinned snapshot of qwen3.7-flash
qwen3.8-max	—	Text	Newest Qwen Max generation
wan2.7-image	N/A	Image, Text, Vision	Image generation/editing, text in images
wan2.7-image-pro	N/A	Image, Text, Video, Vision	Higher-quality image and image-to-video workflows
wan2.7-t2v	—	Video	Text-to-video generation (720p/1080p)
wan2.7-i2v	—	Video	Image-to-video generation (720p/1080p)
wan2.7-r2v	—	Video	Reference-to-video generation (720p/1080p)

DeepSeek

Model ID	Context	Capabilities	Best for
deepseek-r1	262,144	Coding, Text	Reasoning-heavy DeepSeek tasks
deepseek-v3	262,144	Coding, Text	Low-cost general and coding work
deepseek-v3.1	262,144	Coding, Text	Balanced DeepSeek reasoning and coding
deepseek-v3.2	128,000	Coding, Text	Very low-cost current DeepSeek general use
deepseek-v4-flash	262,144	Coding, Text	Very low-cost high-throughput coding/text
deepseek-v4-pro	262,144	Coding, Text	Strong DeepSeek route on chat/completions and messages
deepseek-v4-pro-0813	—	Text	Pinned snapshot of deepseek-v4-pro
deepseek-v4-flash-0731	—	Text	Pinned snapshot of deepseek-v4-flash
deepseek-v4-flash-vision-exp	—	Text	Experimental vision variant of V4 Flash
deepseek-v4.1-flash	—	Text	Newest DeepSeek flash route (V4.1)

Kimi

Model ID	Context	Capabilities	Best for
kimi-k3	—	Text	Newest Kimi generation; Moonshot’s highest-tier route on AIsa
kimi-k2.7-code	—	Text	Coding-focused K2.7 route
kimi-k2.6	128,000	Text	General Kimi reasoning and long-context text
kimi-k2.5	262,144	Coding, Text, Video, Vision	Visual coding, document vision, long-video understanding
kimi-k2-thinking	256,000	Coding, Text	Reasoning and agentic coding

MiniMax and GLM

Model ID	Provider	Context	Capabilities	Best for
MiniMax-M2.5	MiniMax	262,144	Coding, Text	Cost-efficient long-document text and coding
MiniMax-M3	MiniMax	1,000,000	Coding, Text, Video, Vision	1M-context MiniMax with vision/video tags
MiniMax-M2.7	MiniMax	—	Text	Newer MiniMax M2-series route
MiniMax-M2.7-highspeed	MiniMax	—	Text	High-speed variant of MiniMax-M2.7
glm-5	Zhipu GLM	128,000	Coding, Text	Chinese reasoning, bilingual documents, coding
glm-5.1	Zhipu GLM	—	Text	Mid-tier GLM 5 revision
glm-5.2	Zhipu GLM	—	Text	GLM 5.2 revision
glm-5.3	Zhipu GLM	—	Text	Newest GLM 5 revision
glm-5.3-flash	Zhipu GLM	—	Text	Faster GLM 5.3 route
glm-5v-turbo	Zhipu GLM	—	Text	GLM 5V turbo route

ByteDance Seed and Seedream

Model ID	Context	Capabilities	Best for
seed-1-6-250915	262,144	Text, Video, Vision	Stable general Seed route
seed-1-6-flash-250715	262,144	Text, Video, Vision	Fast, low-cost throughput
seed-1-8-251228	262,144	Coding, Text, Video, Vision	Stronger agentic and coding tasks
seed-2-0-mini-260215	262,144	Coding, Text, Video, Vision	Low-cost Seed 2.0 route
seed-2-0-lite-260228	262,144	Coding, Text, Video, Vision	Balanced Seed 2.0 route
seed-2-0-pro-260328	262,144	Coding, Text, Video, Vision	Strongest Seed 2.0 route
seedream-5-0-260128	262,144	Image, Vision	Newer Seedream image route
dreamina-seedance-2-0-260128	—	Video	Dreamina Seedance video generation (up to 4k)
dreamina-seedance-2-0-fast-260128	—	Video	Faster, lower-cost Seedance video route
dreamina-seedance-2-5	—	Video	Newer Dreamina Seedance video route

StepFun, Xiaomi MiMo, and Tencent Hunyuan

Model ID	Provider	Context	Capabilities	Best for
step-3.5-flash	StepFun	—	Text	Low-cost StepFun text route
step-3.5-flash-2603	StepFun	—	Text	Pinned snapshot of step-3.5-flash
step-3.7-flash	StepFun	—	Text	Newest StepFun flash route
mimo-v2.6-flash	Xiaomi	—	Text	Fast Xiaomi MiMo route
mimo-v2.6-pro	Xiaomi	—	Text	Higher-tier MiMo route
mimo-v2.6-pro-ultraspeed	Xiaomi	—	Text	High-speed variant of mimo-v2.6-pro
hy3	Tencent	—	Text	Tencent Hunyuan text route
hy4-preview	Tencent	—	Text	Preview of the next Hunyuan generation

Choosing quickly

Need	Start with
Lowest-cost text or coding	deepseek-v4-flash, qwen-flash, qwen-mt-flash
1M-token Chinese/bilingual context	qwen3.7-max, qwen3.6-plus, MiniMax-M3
Coding agents	qwen3-coder-plus, qwen3.7-max, kimi-k2-thinking, MiniMax-M3
Vision/document/video understanding	qwen3-vl-plus, kimi-k2.5, seed-2-0-pro-260328
Image generation/editing	seedream-5-0-260128, wan2.7-image, wan2.7-image-pro
Chinese-language reasoning	glm-5.2, qwen3.7-max, MiniMax-M3
Video generation	wan2.7-t2v, wan2.7-i2v, dreamina-seedance-2-0-260128
Lowest-cost experimentation	qwen-flash, step-3.5-flash

A — in the Context column means the current gateway metadata does not publish that model’s context window, and the Capabilities column then shows only its base type. Check aisa.one/models for live limits and capability tags.

Provider guides

• Qwen and Wan models
• DeepSeek models
• Kimi models
• ByteDance Seed and Seedream
• MiniMax models
• GLM models
• All supported AIsa models