XingChen AGI API
XingChen (星辰) is the large-model family from China Telecom's AI technology company. Its open Xing4.0 generation is built for agentic work such as task planning, tool calling and multi-step execution, and the 29B A4B variant is listed on Felo as a chat route.
- Models
- 1
- chat ×1
- Input per 1M
- $0
- Output per 1M
- $0
- Largest context
- 262.144K
What XingChen AGI provides
XingChen is China Telecom's model family, developed by China Telecom Artificial Intelligence Technology Co., Ltd. The Xing4.0 generation succeeds the earlier TeleChat series and is trained entirely on Ascend NPUs with the MindSpore framework. Version 29B A4B uses a Mixture-of-Experts design with 29B total parameters and 4B active per token, and the release is published under Apache-2.0.
XingChen's model on Felo
Felo lists Xing4.0 29B A4B as a chat model. It accepts text and returns text, supports tool calling and multi-step planning, and has reasoning enabled by default with the option to turn it off. The route appears in the catalog as a free model, which makes it a low-risk way to test the family before committing to a self-hosted deployment.
Why use XingChen through Felo
A Felo key calls Xing4.0 next to the other chat, image and video models in the catalog, so you can compare it with models from other providers without a separate account. Usage appears on one Felo balance, and the same key stays valid if you later move the workload to another model.
Pricing and access
On Felo, the Xing4.0 29B A4B route is listed as a free model, so calls are not charged per token; a funded Felo balance is still required to send requests, and paid models are deducted from prepaid USD credits. Outside Felo, the weights are published on Hugging Face and ModelScope with vLLM, SGLang and KTransformers support for self-hosting.
Who XingChen AGI is for
Xing4.0 fits agent back ends: task planning, tool use, coding assistance, data analysis and office automation, with Chinese and English input. It is text-only, so image and audio work needs a different model, and reasoning adds latency when it stays on. The open weights also make it a candidate for private deployment: the 4-bit quantized build needs roughly 15GB of VRAM, and a GGUF build runs in tools such as LM Studio with at least 32GB of memory.
Specifications
- Input per 1M
- $0
- Output per 1M
- $0
- Context
- 262.144K
- Max output
- 235.929K
- Protocols
- messages
Similar models from other providers
Frequently asked questions
- What is XingChen AGI?
- XingChen is the model family of China Telecom's AI technology company. Xing4.0 is its open agentic generation, renamed from the earlier TeleChat series and trained on Ascend hardware.
- Which XingChen model can I call through Felo?
- Felo lists Xing4.0 29B A4B as a chat model. The live table on this page shows the current listing and rate.
- Is the XingChen model on Felo free?
- The 29B A4B route is listed as a free model, so token usage is not charged. A funded Felo balance is still needed to send requests, and catalog listings can change.
- Do I need a separate XingChen account?
- No. The model is called with a Felo API key and paid through your Felo balance. Self-hosting the open weights is a separate option if you need private deployment.
- Can I run Xing4.0 locally?
- Yes. The weights are published under Apache-2.0 with vLLM, SGLang and KTransformers support. The 4-bit build needs about 15GB of VRAM, and a GGUF build runs in tools such as LM Studio with at least 32GB of memory.
- Does Xing4.0 support tool calling?
- Yes. It is built for tool calling and multi-step planning, and reasoning can be turned off to cut latency on simpler steps.