AI Briefing#vision-language-model
The latest AI and developer news about #vision-language-model, with the original source and a short summary.
The latest AI and developer news about #vision-language-model, with the original source and a short summary.
Pocket LLM v1.5.0 has been released, adding voice and image input, OCR, and camera capture to Android offline chat.
Running Qwen3.6-27B on a single RTX 3090 achieved 85 TPS at 125K context.
An AI agent analyzed CAD drawings in 10 minutes, replacing 2-3 days of work.
IBM Granite 4.0 3B Vision has been released as a VLM for chart, table, and KVP extraction.
GroundedPlanBench and V2GP, which make VLMs plan actions and locations together, boosted performance.
Qwen3.5-397B-A17B has been released, a native multimodal model in which only 17B of the 397B parameters are activated.
Naver has unveiled an omnimodal AI connecting text, images, and audio with 32B Think and 8B Omni.
By dividing professional video generation into 4 control axes, it exposed the limitations of 10+ models.
Applying LoRA fine-tuning to the Pixtral-12B model dramatically improved satellite image classification performance.
Hugging Face has released nanoVLM, a lightweight toolkit that lets even beginners easily train Vision Language Models.
This summary covers the major technology trends, latest models, and benchmarks in Vision Language Models (VLM) that have rapidly evolved over the past year.
Visual Salamandra, a 7B-parameter multimodal model, has been released with image and video understanding capabilities.
Qwen2.5-Omni, the new flagship model in the Qwen series, is an end-to-end multimodal model that unifiedly processes text, images, audio, and video.
Google has released SigLIP 2, a new vision-language encoder with improved performance and multilingual capabilities.
HuggingFace has released the SmolVLM2 series of ultra-lightweight multimodal models with video understanding capabilities.
Google has launched the PaliGemma 2 Mix models, optimized for a variety of vision-language tasks.
Qwen has released Qwen2.5-VL, its next-generation vision-language model with agentic capabilities and long-video understanding.
HuggingFace's smolagents has been updated with VLM support, enabling it to process visual information.
The Qwen team has released QVQ, an open-weight multimodal model with enhanced visual reasoning capabilities.
Google has unveiled PaliGemma 2, a new vision language model (VLM) built on Gemma 2.
With an average score of 75.1 on Korean benchmarks, surpassing GPT-4o (61.9), the model adopts Native resolution and the C-Abstractor architecture.
Meta has released the Llama 3.2 model series, featuring multimodal capabilities and on-device execution.
Mistral AI released Pixtral 12B, which maintains text performance while offering powerful multimodal reasoning capabilities.
Qwen2-VL, the latest vision-language model in the Qwen model series, has been released with enhanced video understanding and Agent capabilities.