MiMo-V2.5
Xiaomi · multimodal long context
Input (0-256K)
$0.40 / M
Output (0-256K)
$2 / M
Best forMultimodal / text generation
NewDeepSeek-V4-Pro-0813, GLM-5.3-Flash and MiniMax-H3 are liveOne API Key
One API Key
Access 200+ Leading AI Models
Access Qwen, DeepSeek, GLM, Kimi, MiniMax, and more through a single API. Fully OpenAI-compatible, allowing you to switch models without changing your code.
Model ecosystem
Everything You Need for AI Development
Chat, image generation, and video generation in a single platform.
DeepSeekQwenKimiZhipu AIMiniMaxDoubaoHunyuanERNIE BotiFlytek SparkDeepSeekQwenKimiZhipu AIMiniMaxDoubaoHunyuanERNIE BotiFlytek Spark
Language models (LLM)
Chat, reasoning, code generation and long context — the leading models in one API.
DeepSeek-V4Qwen3.7GLM-5.3-FlashKimi-K3DoubaoMiniMax-M3ERNIE Bot...
Image generation
Text-to-image, image-to-image and visual understanding, with high-quality output across styles.
Tongyi WanxiangDoubao SeedreamHunyuan ImageKolorsDeepSeek VisionMiniMax Image...
Video generation
Text-to-video, image-to-video and multimodal generation to produce creative video content fast.
ViduKlingChanjingMiniMax-H3Tongyi Wanxiang VideoDoubao Video...
New Models Available Within Hours
Direct vendor pricing with no intermediary markup, ensuring reliable capacity and competitive rates.
GLM-5.3-Flash
Zhipu AI · general text model
Input
$0.1144 / M
Output
$0.4004 / M
Best forVision / text generation / reasoning
GLM-5.3
Zhipu AI · visual reasoning
Input
$1.144 / M
Output
$4.004 / M
Best forReasoning / vision understanding
DeepSeek-V4-Flash-0731
Cost-effective reasoning
Peak input / output
$0.429 / $1.287
Off-peak input / output
$0.2145 / $0.6435
Best forReasoning / text generation
DeepSeek-V4-Pro-0813
Flagship reasoning and text model
Peak input / output
$1.287 / $3.861
Off-peak input / output
$0.6435 / $1.9305
Best forReasoning / text generation
DeepSeek-V4-Flash-Vision-Exp
Vision understanding and reasoning
Peak input / output
$0.429 / $1.287
Off-peak input / output
$0.2145 / $0.6435
Best forReasoning / vision understanding
MiniMax-M3
Long-context multimodal
0-512K input / output
$0.588 / $2.352
512K-1M input / output
$1.176 / $4.704
Best forMultimodal
MiniMax-H3
Image and video generation
Image input (>5)
$0.0286 / image
Video generation 2K
$0.1144 / sec
Best forVideo generation
Kimi-K3
Moonshot AI · reasoning model
Input
$2.86 / M
Output
$14.3 / M
Best forReasoning / text generation
Qwen3.7-Max
Tongyi flagship reasoning model
Input
$1.716 / M
Output
$2.145 / M
Best forReasoning / text generation
Qwen3.7-Plus
Tongyi multimodal reasoning
Input (0-256K)
$0.28 / M
Output (0-256K)
$1.12 / M
Best forMultimodal / reasoning
Quick start
Get Started in Three Simple Steps
OpenAI-compatible protocol with no code refactoring — from registration to your first call in five minutes.
01
Create a UCloud Global account
Sign up to unlock 200+ leading AI models instantly.
console.ucloud-global.comSign up free
Ready to use
02
Create your API key
Sign in to the console, open the model service platform and generate your API key for authentication and usage tracking.
# Your API Key
sk-••••••••••••••••••Created
03
Configure and call a model
Replace two config lines and set the model ID.Call guide
quick_start.py
# OpenAI SDK compatible
from openai import OpenAI
client = OpenAI(
base_url="https://api.modelverse.cn",
api_key="sk-YOUR_API_KEY"
model="qwen3-max"
)
DeepSeek-V3.2 Qwen3-Max GLM-4.6 Kimi-K2 Doubao 200+
Why choose AstraFlow?
Enterprise-Grade Platform Backed by a Public Company
Directly connected to model providers with the negotiating power to pass original manufacturer volume discounts to you.
New Models Available Within Hours
DeepSeek, Qwen, GLM and other models are delivered through compliant, licensed channels.
Consistent Performance During Peak Demand
Traffic is intelligently distributed across multiple provider channels to maintain performance during peak demand for high-volume users, with dedicated resource pools available.
Enterprise-grade security and compliance
UCloud Global does not train foundation models and never uses customer data for model training.
Dedicated One-on-One Support
A 24/7 technical team on call, with deployment and integration guides and hands-on support.
app.py
utf-8
from openai import OpenAI
# Use AstraFlow exactly like OpenAI
client = OpenAI(
api_key="sk-AstraFlow-xxxxxxxx",
base_url="https://api.modelverse.cn"
)
# Call Qwen or DeepSeek at full speed
response = client.chat.completions.create(
model="qwen-2.5-72b-instruct",
messages=[
{
"role": "user",
"content": "Tell me about UCloud Global"
}
],
stream=True
)
for chunk in response:
print(chunk.choices[0].delta.content or "", end="")
Fully compatible with the OpenAI SDKLn 12, Col 4
FAQ
Frequently asked questions
What is AstraFlow?
AstraFlow is UCloud Global's AI model gateway. It brings together 200+ leading models from DeepSeek, Qwen, GLM, Kimi, Doubao and others behind a single API key. The platform is OpenAI-compatible, so developers can switch models without refactoring existing code, across chat, image generation, and video generation.
Which models does AstraFlow support?
AstraFlow connects 200+ models from 15 vendors, including BAAI, ERNIE Bot, Bilibili, ByteDance, DeepSeek, EasyLink, HappyHorse, Kuaishou, MiniMax, Moonshot AI, Qwen, Vidu, Zhipu AI, Xiaomi and Chanjing — spanning language models, image generation and video generation, including the DeepSeek series, the Qwen3 series and GLM-5.3. New models are typically synced within hours of release.
Which SDKs and protocols are supported by AstraFlow?
AstraFlow is fully compatible with the OpenAI API protocol and works with the OpenAI Python SDK, LangChain, LlamaIndex and other mainstream frameworks. Point base_url at https://api.modelverse.cn, add your API key and set the model ID. Migrating from OpenAI typically requires changing only two lines of code.
How does AstraFlow protect my data and privacy?
The platform is built on UCloud Global's cloud infrastructure. All data is encrypted in transit, AstraFlow does not retain customer prompts or request content, and user data is never used for model training. Enterprise plans also support private deployment and calls over a VPC, meeting enterprise requirements for data security and compliance management.
How is AstraFlow billed?
AstraFlow is usage-based: you are billed per token, at rates that differ by model, with no minimum spend. Billing can be settled daily or monthly, and enterprise customers can contact sales for dedicated discounts and annual or monthly packages.
How soon after release is a new model available on AstraFlow?
AstraFlow works closely with every model vendor, so mainstream large models are typically synced within hours of release — DeepSeek-V3.2 and the Qwen3 series, for example, were available shortly after their official release. New-model notices appear in the Model Center and in announcements, and you can also query the current list of available models through the API's models endpoint.
What is the difference between AstraFlow and the UCloud Global console?
The UCloud Global console is the general-purpose platform for cloud resources such as compute, storage and networking. AstraFlow (ModelVerse) is the sub-platform dedicated to model services. AstraFlow is available through the Model Center module within the UCloud Global Console. In AstraFlow you create API keys, browse the model list and manage usage and cost, while cloud resource management still happens in the main console.
$5CREDIT
Link a credit card to unlock Alipay, WeChat Pay & SEA wallets.
Contact support to get a $5 credit. Need help? We’ve got you.
Ready to build your AI application?
Trusted by 1,000+ enterprise developers worldwide. New accounts can claim free usage credits.
