Tencent releases WeVisOne-VL-4B multimodal model for local use

Tencent has released WeVisOne-VL-4B, a multimodal perception model built on Qwen3-VL, capable of tasks like object detection, segmentation, and OCR. This 4B parameter model can be run locally on an 8GB GPU using Ollama or llama.cpp, though mask generation requires a separate decoder. Tencent's benchmarks show WeVisOne-VL-4B outperforming Qwen3-VL-8B and other models on specific tasks like matting and rejection of absent targets, though the comparison is noted to be missing some key baselines. AI

IMPACT Enables local deployment of advanced multimodal perception capabilities on consumer hardware.

RANK_REASON The item describes how to run a specific multimodal model locally, focusing on technical implementation details and performance on consumer hardware.

Read on dev.to — LLM tag →

• arXiv:2610.31329
• LISA-13B
• llama.cpp
• Ollama
• Qwen3.5-9B
• Qwen3-VL
• Qwen3-VL-8B
• Rex-Omni-3B
• SAM2-Matting
• Tencent
• WeChat Vision
• WeVisOne-VL-4B

• model release
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →