On June 28, AXera-Chip organized a Developer Ecosystem Salon event in Shenzhen. Tang Qi, who leads the open-source community solutions team at AXera-Chip, disclosed that by June 2026, the AXera-Chip platform model library had amassed over 200 models. These models span across a wide spectrum, including Large Language Models (LLMs), Vision Language Models (VLMs), multimodal models, and specialized vertical detection models. Notably, the library has been seamlessly integrated into two prominent model communities: Hugging Face and ModelScope. Moreover, the newly developed Pulsar2 6.0 toolchain has undergone a comprehensive iteration, ensuring compatibility with the entire range of chips while significantly enhancing the inference performance of mainstream on-device large models, such as Qwen and MiniCPM. Additionally, the company's next-generation NPU architecture, tailored specifically for large models, has demonstrated a marked improvement in performance, boasting native support for popular frameworks like PyTorch, Sglang, and vLLM.