AI Briefing#multimodal
The latest AI and developer news about #multimodal, with the original source and a short summary.
The latest AI and developer news about #multimodal, with the original source and a short summary.
This summary covers the major technology trends, latest models, and benchmarks in Vision Language Models (VLM) that have rapidly evolved over the past year.
Kakao has released 'Kanana-o', a multimodal LLM that integrates its voice model (Kanana-a) and vision model (Kanana-v) using model merging techniques, demonstrating superior performance compared to global models in Korean speech recognition (CER 8.5).
Meta has released Llama Guard 4, a 12B-scale multimodal safety model that inspects both text and images.
OpenAI has released gpt-image-1, a model that provides ChatGPT's multimodal capabilities through the API.
OpenAI unveiled o3 and o4-mini, which perform visual reasoning by directly cropping, zooming, and rotating images.
Cohere's major LLM and multimodal models are now available for serverless inference on the Hugging Face Hub.
Visual Salamandra, a 7B-parameter multimodal model, has been released with image and video understanding capabilities.
Meta's next-generation MoE models, Llama 4 Maverick and Scout, have been released on Hugging Face.
OpenAI has integrated sophisticated, natively multimodal image generation capabilities into GPT-4o.
GPT-4o delivers native image generation with image creation and transformation capabilities more powerful than DALL·E 3.
Mistral released Mistral Small 3.1, an open-source model with multimodal capabilities and a 128k context.
Google has released Gemma 3, a new open model supporting multimodality and a 128k context.
Mistral has launched the new Mistral OCR API, which accurately extracts text, tables, formulas, and more from complex documents.
Cohere For AI has released the 'Aya Vision' series, an open-weight multimodal model supporting 23 languages.
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.
HuggingFace has released SmolVLM 256M and 500M, ultra-lightweight multimodal models.
An embedding model for multilingual visual document retrieval, developed in collaboration with LlamaIndex, along with a 500K dataset, has been released.
xAI has introduced a new image generation model, Aurora, into Grok and launched it on the 𝕏 platform.
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.
HuggingFace has released SmolVLM, a 2B-scale open-source vision language model that maximizes memory efficiency.
Kakao has unveiled the 'Kanana' family of AI models specialized for Korean language processing, demonstrating overwhelming performance in Korean compared to global SOTA models and already applied to services such as KakaoTalk.
xAI has launched a public beta of its API for using Grok models and is providing $25 in free credits every month.
OpenAI has launched vision fine-tuning, a feature that lets developers train GPT-4o using both images and text together.