MiniCPM-V 4.7
A multimodal vision-language model released by OpenBMB. It uses a SigLIP vision encoder together with a Qwen3.5 language model backbone, and supports multiple visual downsampling modes.
Timeline (last 90 days)
OpenBMB's MiniCPM-V 4.7 continues the combination of a SigLIP vision encoder plus a Qwen3.5 language model backbone, supports two visual downsampling modes, 4x and 16x, and adds canvas M-RoPE to lay images on a 2-D canvas to assign (temporal, height, width) positions to visual tokens.
The author used MiniCPM-V 4.7 to complete all development and testing of pricehunt; the extension supports any vision model running via Ollama, or any server compatible with the OpenAI interface.
MiniCPM-V 4.7 supports 4x and 16x hybrid visual token compression, allowing developers to flexibly switch compression ratios according to task needs.
MiniCPM-V 4.7 continues the MiniCPM-V family's single-image, multi-image, and video understanding capabilities, with foundational capabilities enhanced compared to the previous generation.
MiniCPM-V 4.7 supports the three major mobile platforms iOS, Android, and HarmonyOS, and provides open-source on-device deployment solutions.
ModelBest (OpenBMB) released MiniCPM-V 4.7, the newest member of the MiniCPM-V series, including 35B-A3B, the series' first MoE architecture variant.
MiniCPM-V 4.7 is compatible with the SGLang, vLLM, llama.cpp, and Ollama inference frameworks, and supports the SWIFT and LLaMA-Factory fine-tuning frameworks.
MiniCPM-V 4.7's quantization formats cover GGUF, BNB, AWQ, and GPTQ, and it adds FP8 quantization support, with the GGUF version already uploaded to Hugging Face.
MiniCPM-V 4.7 provides day-0 support for Rockchip, Houmo, Intel, AMD, and Arm platforms.
On the vision side, MiniCPM-V 4.7 adopts the early compression inside the ViT from LLaVA-UHD v4.
All Facts (10)
OpenBMB's MiniCPM-V 4.7 continues the combination of a SigLIP vision encoder plus a Qwen3.5 language model backbone, supports two visual downsampling modes, 4x and 16x, and adds canvas M-RoPE to lay images on a 2-D canvas to assign (temporal, height, width) positions to visual tokens.
50%UnverifiedThe author used MiniCPM-V 4.7 to complete all development and testing of pricehunt; the extension supports any vision model running via Ollama, or any server compatible with the OpenAI interface.
50%UnverifiedMiniCPM-V 4.7 supports 4x and 16x hybrid visual token compression, allowing developers to flexibly switch compression ratios according to task needs.
50%UnverifiedMiniCPM-V 4.7 continues the MiniCPM-V family's single-image, multi-image, and video understanding capabilities, with foundational capabilities enhanced compared to the previous generation.
50%UnverifiedMiniCPM-V 4.7 supports the three major mobile platforms iOS, Android, and HarmonyOS, and provides open-source on-device deployment solutions.
50%UnverifiedMiniCPM-V 4.7 is compatible with the SGLang, vLLM, llama.cpp, and Ollama inference frameworks, and supports the SWIFT and LLaMA-Factory fine-tuning frameworks.
50%UnverifiedMiniCPM-V 4.7's quantization formats cover GGUF, BNB, AWQ, and GPTQ, and it adds FP8 quantization support, with the GGUF version already uploaded to Hugging Face.
50%UnverifiedMiniCPM-V 4.7 provides day-0 support for Rockchip, Houmo, Intel, AMD, and Arm platforms.
50%UnverifiedOn the vision side, MiniCPM-V 4.7 adopts the early compression inside the ViT from LLaVA-UHD v4.
50%UnverifiedModelBest (OpenBMB) released MiniCPM-V 4.7, the newest member of the MiniCPM-V series, including 35B-A3B, the series' first MoE architecture variant.
50%