May 14th, yesterdayWall-facing intelligenceOfficially released andOpen Source 1.3B SIDEMultimodal large model MiniCPM-V 4.6, provides two versions of Instract and Thinking。
Instruct version: missions for common image understanding, STEM mathematical reasoning, OCR documentation, video time-series understanding and target positioning, going beyond Qwen3.5-0.8B and Gemma4-E2B-it
Thinking version: High-level tasks such as multi-image association reasoning, hallucinating inhibition, are also ahead of the same size。
It was described that the token topping volume based on vLLM was 1.5 times greater than Qwen3.5-0.8B, and that the consumption of token in AA was only about 1/19 for the latter's non-deprecative version; and that the initial delay of the high-clean image was 75.7 ms, 2.2 times faster than Qwen3.5-0.8B。
💻 GitHub: github.com/OpenBMB/ MiniCPM-V
Hugging Face: huggingface.co/openbmb/ MiniCPM-V-4.6
👾ModelScope: modelscope.cn/models/OpenBMB/ MiniCPM-V-4.6