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.
The FP8-quantized release targets reduced repetitive reasoning and improved tool adherence for long-horizon tasks.
Built by contrastively fine-tuning meta-models/Muse-Glimmer-30B, the model supports arbitrarily large candidate sets through ANN indexing.
The guide maps Claude Code's default model IDs to the 560B-total, 25B-active multimodal MoE via an Anthropic-compatible interface.
The 560B-parameter model supports text, image, and video inputs and achieves 92.21 on OmniDocBench 1.5.
Kakao has released the 3B-scale multimodal decision-making model Kanana-2-v-decide as open source.
The summary introduces three major trends—expanding agent evaluation scope, structured learning, and refining data quality—along with 10 papers.
The model uses a 255-way readout head to output probabilities over candidate answers in a single forward pass, avoiding token generation entirely.
The model uses a looped decoder to execute 56 layer passes from 32 distinct layers, supporting text and image inputs for agentic workflows.
The new models use full-image inference without zooming tools, improving overall accuracy from 55.52 to 60.51 across six domains.
Liquid AI has released six 2B-scale LFM3 models supporting execution on edge devices and multimodal processing.
The method uses self-generated question-answer pairs and reasoning traces to boost performance across text and multimodal domains while retaining capabilities in untrained areas.
NVIDIA releases five Nemotron 3.5 models for embedding, reranking, and parsing, with the 8B Embed model topping ViDoRe v3.1 and the suite reducing agent token costs by up to 14%.
Kanana-2-vision-encoder recorded the global highest scores in 8 out of 9 domains.
The 180B parameter hybrid attention MoE model features a 1,048,576-token context window and selectable reasoning effort levels.
The compact Mixture-of-Experts model supports speech recognition, audio reasoning, and optional synthesized speech responses in Chinese and English.
DigUp is a desktop app released under the MIT license that uses EmbeddingGemma 2 to search photos, PDFs, videos, and code locally without the cloud.
The new multimodal decision models, Decider V1.2 27B and Decider Flash V1.2 4B, classify content and route requests via typed questions.
The model supports multimodal retrieval across 34 languages for text and 18 for images, with open commercial use rights.
NVIDIA released Nemotron 3.5 Super VL, a 120-billion-parameter open model, alongside two edge-optimized Bolt models for local agentic workloads.
The multimodal model returns structured probabilities for classification and routing tasks rather than generating text.
The model features 42B active parameters per token, a 1M-token context window, and supports text, image, audio, and video inputs.
The new Apache-2.0 model family, ranging from 0.6B to 27B parameters, achieves state-of-the-art scores on both text and vision boards of the Jev Decision Index.
Microsoft has released AesCode 8B and 32B, models that generate editable HTML/CSS visual artifacts like slides and dashboards by using image references for layout while preserving semantic accuracy.
Perplexity has released pplx-embed-v2-late, a family of ColBERT-style multi-vector embedding models in 0.6B and 9B sizes, supporting multimodal retrieval for text and visual documents.