XingChen-AGI has released Xing4.0-29B-A4B, a 29 billion parameter language model with 4 billion active parameters per token. This model boasts a 256K context length, expandable to 512K, and is optimized for complex engineering tasks. It features an agent-oriented architecture supporting multi-step planning and tool calling, and was trained entirely on the Ascend NPU platform using the MindSpore framework, achieving significant training efficiency gains. AI
IMPACT This model's advanced agent architecture and long context capabilities could accelerate development in complex reasoning and multi-step task execution for AI applications.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Ascend NPU
- Claude Code
- Hermes
- Hugging Face
- KTransformers
- LLaMA-Factory
- MindFormers
- MindSpore
- OpenClaw
- OpenCode
- SGLang
- vLLM
- Xing4.0-29B-A4B
- XingChen-AGI
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →