MODEL ACCESS GUIDE
Access selected Chinese AI models through one API layer
Use consistent authentication and operational controls while choosing models by capability, availability, latency, and cost.
Updated 2026-09-10
TokenAAS lists selected DeepSeek, Qwen, GLM, Kimi, MiniMax, and Doubao model families and exposes configured production routes through public model IDs.
Choose by workload
Do not choose only by vendor name. Compare reasoning quality, context requirements, latency, modality, and the current price shown in the catalog.
- DeepSeek for general and reasoning workloads
- Qwen and GLM for broad Chinese-language use cases
- Kimi and MiniMax for additional text options
- Doubao for text, image, and video workflows where configured
Separate catalog from routing
A catalog entry may be Available, capacity-limited, unavailable, disabled, or Coming soon. Only routable models should receive production traffic.
Design a fallback policy
For critical workloads, define an approved fallback model with compatible output expectations instead of switching providers blindly.
Frequently asked questions
Does TokenAAS expose upstream provider credentials?
No. Applications authenticate with their TokenAAS API key; upstream credentials remain part of the managed routing configuration.
Are all listed models immediately callable?
No. Check the status badge and verify the model appears in GET /v1/models for the intended API key.