Search “Multimodal”

Articul8 Launches Multimodal Decision Model Bivia

Articul8 has released the multimodal decision model Bivia, which can process documents, images, audio, video, and time-series data, make continuous decisions, and defer to human judgment when uncertain. Articul8 grew out of a project delivered to enterprise customers by an Intel team one month before the launch of ChatGPT; the company is valued at $500 million and says it has accumulated more than 3 years of production deployment experience in industries including semiconductors, energy, and manufacturing.

Google Releases EmbeddingGemma 2 Paper: A 744M-Parameter On-Device Multimodal Embedding Model

Google has released the EmbeddingGemma 2 paper, an Apache 2.0-licensed on-device multimodal embedding model that maps text, code, images, audio, and video into the same vector space. The model is built on Gemma 4 and tuned through contrastive training by learning from the larger Gemini Embedding 2; with 744M parameters, it surpasses models 2 to 12 times larger on multiple tasks. On the Pixel 11 Pro, the quantized text weights occupy about 191MB of active memory, while the full multimodal model takes about 567MB; the context window is expanded 4 times to 8K tokens, accommodating about 5.5 minutes of audio, 29 images, or 58 video frames; directly embedding speech scores 69.54 on the MSEB spoken query retrieval benchmark, higher than the 57.59 of transcribing first and then retrieving, and code retrieval on MTEB Code improves from 68.76 in EmbeddingGemma 1 to 78.68.

Mistral Opens the Weights of Mistral Large 4, a 1-Trillion-Parameter Natively Multimodal Model

Mistral AI announced that it is opening the weights of its frontier model Mistral Large 4 (codename Le Chonk), which has 1 trillion parameters with 52 billion active parameters and native multimodal support. ML4 was trained in a European data center on 3,800 Nvidia Grace Blackwell GPUs, improving from 28% to 41% on Terminal-Bench 4.0, tying GLM-5.3 for the best coding-agent score while using an average of 47% fewer tokens per task. Open weights mean enterprises can deploy it on their own hardware and behind their firewalls, and the model offers three reasoning modes: low, medium, and high.

Multimodal Retrieval Engine Selection: Applicability Conditions and Solution Design for EMR Serverless StarRocks

EMR Serverless StarRocks unifies full-text search, scalar filtering, and vector similarity search into a single execution engine, completing hybrid recall, RRF fusion reranking, and lake table column-level update write-back with a single SQL statement. It is suitable for multimodal data scenarios organized with open lake formats such as Paimon/Iceberg/Lance and that rely heavily on SQL interaction; if a RESTful vector service interface or millisecond-level high-throughput point queries are required, an additional gateway wrapper is needed or a dedicated vector database should be considered.

A Beauty Company's FDE Implementation in Practice: An Eight-Step Pipeline for Cleaning Massive Multimodal Data into Trustworthy Data

A beauty company's FDE implementation project processes text, image, audio, and video data collected from Xiaohongshu, Douyin, WeChat Channels, and Weibo through eight steps: defining data specifications, preserving original content, mapping data relationships, multimodal normalization, conflict and quality cleaning, generating structured data, performing acceptance checks, and ingesting according to rules and versions. The original article emphasizes that cleaning results must not overwrite the original text, and that OCR, ASR, and model labels are only added results; it also describes the practice of tracing back to the original evidence when ASR and OCR values conflict, and placing cases that cannot be confirmed into a pending-verification queue. This pipeline has also been organized by the author into a general solution reusable in scenarios such as consumer goods insights, automotive word-of-mouth, restaurant site selection, and brand public opinion.

Tianjin University's Zhang Changqing introduces generalized multimodal fusion at CNCC 2026: from trustworthy evidence fusion to the Fusion Anything foundation model

At CNCC 2026, Tianjin University's Zhang Changqing presented the report "Generalized Multimodal Fusion Technology," introducing the team's research on multimodal representation, dynamic trustworthy fusion, and the general-purpose prediction foundation model Fusion Anything Model (FAM). The TMC method uses evidence theory to estimate uncertainty, raising crystal structure recognition accuracy from about 58%–59% to around 93.9%. FAM was tested on 18 real-world datasets, 12 modalities, and 11 types of prediction tasks, achieving an average accuracy about 6.9 percentage points higher than specialized fusion baselines.

Google Releases Genie 4, a Real-Time Multimodal World Model

Google has released Genie 4, adding native audio, movable and talking characters, more refined physics, and real-time prompt editing. Genie is a real-time multimodal world model that can render playable environments frame by frame based on keyboard operations and text prompts. A waitlist is currently open, with priority for Google AI Ultra users.

Zhipu May Launch Native Multimodal Model GLM-Omni

X.PIN claims that Zhipu may launch a native multimodal model, GLM-Omni. This system combines image generation with text training, having the model draw out visual reasoning steps before answering, raising text question-answering accuracy from 12.57% to 41.14%. The author also jokes that this is like doodling during one's own exams.

English translation is being prepared.

Flatkey Gateway Lets Codex/Claude Code Call Multimodal Models

Flatkey, as a unified API gateway, lets Codex and Claude Code call text models such as GPT, Claude, DeepSeek, GLM, and Kimi as well as image and video models such as GPT Image and Seedance via CLI. One Key provides access to 100+ official models and 1,000+ tool APIs, with costs consolidated into a single bill. Everyday prices are at least 20% lower than official pricing. During the October 19–23 promotion, GPT, GLM 5.2, Kimi K3, and DeepSeek are 40% off, Claude is 20% off, and Seedance 2.0 is 15% off.

Alibaba Cloud OSS Three-Bucket-in-One Architecture, Multimodal Retrieval and Lifecycle Management, and ossutil Hands-On Tutorial

Alibaba Cloud OSS has launched a "object bucket + vector bucket + table bucket" three-bucket-in-one architecture that unifies the storage of unstructured, vector, and structured data. The vector bucket supports trillion-scale vectors and millisecond-level similarity retrieval, and the table bucket is compatible with Apache Iceberg. The platform offers five storage tiers and lifecycle rules to automatically tier down data, and includes ossutil command-line installation and configuration, upload and sync, and Python SDK examples.

English translation is being prepared.

English translation is being prepared.

Central China Normal University proposes CoReFuse-Med: purify first, then fuse, to address low-quality modalities dragging down multimodal MRI segmentation

A team from Central China Normal University and others proposed the multimodal MRI segmentation network CoReFuse-Med, which uses LACE+IMSH to remove resampling artifacts at shallow layers and MBCF to calibrate modality contributions at deep layers, solving the problem of low-quality modalities contaminating fusion. On the EPVS dataset, it achieves a DSC of 0.7326 and an HD95 of 7.35, outperforming MedNeXt's 0.7085 and 8.17; on BraTS, the DSC is 0.8528, with only 2.665M parameters. This method targets the resolution degradation after spatial alignment and does not cover scenarios with missing modalities or strong noise.

English translation is being prepared.

Luma Proposes Exact Sequence Parallelism for Packed Multimodal Masks

Luma introduces an exact sequence parallelism method for packed multimodal masks: each layer chooses among ring attention, sparse all-to-all, or Ulysses, with numerical results consistent with a single GPU, and on up to 128 H100s it outperforms the FlashMask + Ulysses hybrid scheme in 47 of 48 cases.

English translation is being prepared.

English translation is being prepared.

Perplexity Open-Sources pplx-decider-v1.2 and Flash Multimodal Decision Models

Perplexity has released its next-generation open-weight multimodal decision models, comprising two sizes, pplx-decider-v1.2 and pplx-decider-flash-v1.2, with improvements in both text and vision scores, inference up to 5x faster, and all questions in a single request completed in one pass. Hugging Face co-founder Clément Delangue reposted the news and noted that the weights have been published on Hugging Face. Perplexity says the flash version is the top-ranked 4B model on the Decision Index.

English translation is being prepared.