Chinese AI API Aggregator with one key for multiple providers
A typed decision API for routing, classification, and scoring
Stage
Not given
Not given
Founded
Not given
Not given
Based in
Not given
CA
Pricing model
Paid
Paid
Pricing
No platform subscription fee. Text models billed per 1M tokens (input, output, and cached reads counted separately). Image models billed per request. Minimum top-up is $10. Example rates: Kimi K3.1 Pro $8/M tokens, DeepSeek V4.1 Flash 1017 $0.836/M tokens, MiniMax M3.1 Flash free until Oct 31 2026,
$0.040 USD per million input tokens; output is free
Key features
Access to DeepSeek, Kimi, GLM, Omen Alpha, Ox Alpha, and other Chinese AI models
OpenAI-compatible API for text models
Unified API key for multiple model providers
Live pricing and model directory
Usage tracking and billing by model and workload
Support for text, image, video, and audio models
Global endpoints for access outside China
Asynchronous workflows for media models
Model directory and live console for configuration
Developer documentation and API guides
Choice question type with multiple options and probability scores
Boolean question type for true/false judgments with probability
Score question type with min/max range and level probabilities
Filter list endpoint for ranking JSON items by instruction match
Support for image and audio input with Crossroads-M model
Idempotency keys for safe retries without duplicate charges
Rate limiting with 120 requests per minute default and 4 concurrent requests
Multiple question types per request with single round trip
What makes it different
Single API key to access multiple Chinese AI providers without managing separate credentials
Transparent, live pricing with real-time model availability shown in console
OpenAI-compatible interface for easy model switching
Usage-based billing with no platform subscription fee
Global access to Chinese AI models from outside China
Returns probability scores for every answer to guide decision-making confidence
Supports image and audio analysis alongside text with dedicated audio-trained model
Filter lists by instruction match without writing text generation code
Typed output constrains format while preserving flexible judgment logic
TokenRa
Crossroads