Xing4.0 29B - Free MoE Model on SiliconFlow, 256K Context
XingChenAGI/Xing4.0-29B
Model Information
| Property | Value |
|---|---|
| Context length | 262,144 tokens (native 256K, extensible to 512K) |
| Input price | ¥0 / 1M tokens |
| Output price | ¥0 / 1M tokens |
| Provider | China Telecom TeleAI / SiliconFlow |
| Status | Free |
| Input modalities | Text |
| Total parameters / active parameters | 29B / 4B |
| Release date | 2026-09-17 |
Description
Xing4.0-29B-A4B is an open-source mixture-of-experts model from the Institute of Artificial Intelligence at China Telecom (TeleAI), and the next generation of the TeleChat series. With 29B total parameters and only 4B activated per token, it natively supports a 256K context that can be extrapolated to 512K, and it is optimized for multi-step planning, tool calling, and execution of complex reasoning chains.
The model was trained entirely on Ascend 910C NPUs with the MindSpore framework, using an mHC + MLA + MTP architecture with 64 routed experts and 1 shared expert, 40 layers in total, and a hidden dimension of 3584. It supports fine-tuning with LLaMA-Factory and MindFormers, as well as inference frameworks such as vLLM, SGLang, and KTransformers, making it well suited to agentic coding, long-text processing, and enterprise private deployments.
Usage Example
curl https://api.siliconflow.cn/v1/chat/completions \
-H "Authorization: Bearer $SILICONFLOW_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "XingChenAGI/Xing4.0-29B",
"messages": [{"role": "user", "content": "Help me draft a multi-step code refactoring plan"}]
}'Related Links
📚 Related Free Models
These models are also available for $0 on SiliconFlow. Choose by context length and modality.
| Model | Context | Description |
|---|---|---|
| Hunyuan-MT-7B | 33K | Hunyuan-MT-7B is Tencent Hunyuan’s lightweight 7B |
| PaddleOCR-VL 1.5 | 16K | Baidu’s 0.9B multimodal document parsing model, with 94.5% |
📋 View all SiliconFlow free models →