OpenBMB has released the MiniCPM-V 4.7 series of multimodal large language models, featuring both a 35B-A3B Mixture-of-Experts (MoE) variant and a 1B variant. The 35B-A3B model is notable for being the first open-source MoE model capable of smooth on-device inference, achieving over 13 tokens/s on an iPhone 17 Pro Max using its Lookahead Expert Scheduling (LES) technique. Both models offer improved performance and efficiency, support broad platform and chip compatibility, and are developer-friendly with support for various inference and fine-tuning frameworks. AI
IMPACT Sets new benchmarks for on-device multimodal inference, particularly for MoE architectures.
RANK_REASON Model release from a notable AI lab (OpenBMB) with specific technical innovations.
Read on Hugging Face Trending Models →
- Google Colab
- Hugging Face
- Kaggle
- openbmb/MiniCPM-V-4.7-35B-A3B
- SGLang
- transformers
- vLLM
- MiniCPM-V 4.7
- MiniCPM-V 4.7-1B
- MiniCPM-V 4.7-35B-A3B
- OpenBMB
- Qwen3.5 0.8B
- Qwen3.6 35B-A3B
AI-generated summary · Google Gemini · from 4 sources. How we write summaries →