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.
Google has unveiled Gemma 4 12B, a new 12B multimodal model that processes vision and audio directly without encoders.
MiniMax has launched the M3 model featuring a 1 million-token context and multimodal capabilities, with weights to be released within 10 days.
For cost efficiency in RAG systems, this proposes storing images by describing them as text at the indexing stage.
NVIDIA has unveiled Cosmos 3, an omnimodal world model for physical AI research.
Google unveiled the video generation model Gemini Omni and Gemini 3.5 Flash with enhanced agentic capabilities.
StepFun has released Step 3.7 Flash, a 196B-scale MoE-based multimodal model.
Cohere has released Command A+, an open-source MoE model that ensures data sovereignty for enterprises and the public sector.
Google has officially launched the image generation models Nano Banana 2 and Pro, strengthening support for enterprise creative workflows.
MiniMax plans to release M3, a multimodal model using Deepseek-style attention, as an open weight model.
A 65k-scale hybrid router that efficiently switches between local and cloud models has been unveiled.
Keye-VL-2.0-30B-A3B, the latest 30B-class multimodal model in the Keye series, has been released.
NuExtract3, a 4B-scale vision-language model specialized in document understanding and structured information extraction, has been released.
LangChain unveiled a new architecture that goes beyond token streaming to stream messages, tool calls, and sub-agents as structured events
LiteFrame is a lightweight vision encoder that cuts Video LLM inference latency by 35% and enables long-form video understanding.
Cohere has open-sourced Command A+, a 218B-parameter MoE model, under the Apache 2.0 license.
Google has redesigned its search box for the first time in 25 years, declaring a paradigm shift toward AI agent-centered search.
The webui of llama.cpp has added the ability to use video files as input.
A template error in the Qwen3.5 Vision model has been fixed in vLLM.
InternLM has released Intern-S2-Preview, a 35B scientific multimodal model.
Amazon Bedrock has released a tool that helps optimize prompts and switch models by comparing up to 5 models.
Promptimus automatically improves already-good LLM prompts without manual work.
Thinking Machines has released a research preview that handles audio, video, and text in real time.
SenseNova has released U1 A3B, a multimodal model based on NEO-unify.
ElevenLabs' **ElevenAgents** now supports new modalities such as images, files, and location information, strengthening its end-to-end problem-solving capabilities.