Vivo's bluelm model leads edge AI performance benchmark

SuperCLUE recently released the latest on-device mobile large model evaluation rankings, highlighting the performance of smartphone-side AI models. vivo's BlueLM3.5Nano3B model secured the top position in the overall ranking with a score of 89.86. This 3B parameter model not only led in the edge-side category but also achieved scores approaching those of major cloud-based large models such as Google's Gemini3.6Flash, Doubao Seed2.1pro, and Qwen3.8Max. In the edge-side rankings, Qwen3.59B and Qwen3.54B scored 87.82 and 85.16 respectively, placing second and third. Other participants included GLM4.6V Flash, Gemma4E4B it, and Gemma4E2B it. The growing adoption of edge-side AI is making the advantages of local operation more evident. Edge-side large models enable image and text reasoning directly on the phone without requiring a constant internet connection. All user input remains on the device, offering a more secure and private experience. Despite hardware limitations, edge-side models cannot scale infinitely like cloud-based models and still face challenges with ultra-complex tasks. However, vivo's achievement with a 3B parameter model demonstrates that small-volume edge-side models can reach a high technical ceiling. This suggests future smartphone AI experiences may rely less on network connectivity, enabling smooth operation even in offline environments. While benchmark scores reflect laboratory performance, actual user experience depends on deep system optimization by major smartphone manufacturers.