Quanta AI Labs

01 / Dec 29, 2025 – Jan 4, 2026 · 20 RELEASES

AI News Week 1, 2026

Yume 1.5 and SAM 3D advance world/3D generation, ROMA and LLMRouter enable agent infrastructure, with SpotEdit and HiStream pushing efficient video editing.

Language Models (2)

  • Youtu-LLM-2B

    Tencent · Small Language Models

    Efficient 2B parameter language model optimized for specialized tasks with vLLM deployment support.

  • Luxical-One

    Datology AI · Text Generation

    Advanced language model with optimized training and inference capabilities.

Vision & Image (2)

  • SpotEdit

    NUS · Image Editing

    Training-free selective region editing in Diffusion Transformers that processes only modified regions.

  • TwinFlow-Z-Image-Turbo

    Inclusion AI · Text-to-Image

    One-step generation on large models with self-adversarial flows for fast image synthesis.

Video & Animation (5)

  • Stream-DiffVSR

    Research · Video Editing

    Low-latency streamable video super-resolution via auto-regressive diffusion for real-time enhancement.

  • ProEdit

    iSEE Lab · Video Editing

    Inversion-based image and video editing from prompts with improved source information handling.

  • HiStream

    Research · Text-to-Video

    Efficient high-resolution video generation via redundancy eliminated streaming with temporal and spatial compression.

  • JavisGPT

    Javis Verse · Video Understanding

    Unified multi-modal LLM for sounding-video comprehension and generation with audio-visual understanding.

  • Space-Time Pilot

    Research · Text-to-Video

    Generative rendering of dynamic scenes across space and time with AR demos and Cam×Time dataset.

3D & Spatial (3)

  • SAM 3D

    Meta AI · 3D Reconstruction

    Bring any 2D image to life with accurate 3D reconstruction of objects and humans including shape and pose.

  • UltraShape 1.0

    PKU Yuan Group · 3D Generation

    High-fidelity 3D shape generation via scalable geometric refinement with advanced mesh quality.

  • Yume 1.5

    Research · 3D Generation

    Text-controlled interactive world generation with keyboard exploration, long-video support, and event control.

Agents & Automation (1)

  • ROMA

    Sentient AGI · Agent Framework

    Recursive Open Meta-Agent framework for building high-performance multi-agent systems with modular architecture.

Developer Tools (4)

  • IQuest Coder

    IQuest Lab · AI Infrastructure

    AI coding assistant platform for enhanced developer productivity and code generation.

  • Model Memory Calculator

    Kolosal AI · AI Infrastructure

    Browser-based GGUF metadata reader for calculating model memory requirements.

  • LLMRouter

    UIUC ULab · AI Infrastructure

    Open-source library for LLM routing with single-round, multi-round, personalized, and agentic routers.

  • llm-d

    LLM-D Project · Inference Engine

    Achieve SOTA inference performance with modern accelerators on Kubernetes for LLM deployment.

Research & Safety (3)

  • App Ideas Collection

    Community · Survey & Overview

    Curated collection of 100+ application ideas across beginner, intermediate, and advanced tiers.

  • Circuit Sparsity

    OpenAI · Research Paper

    Sparse model research from Gao et al. 2025 exploring efficient neural network architectures.

  • Qwen-Image ComfyUI Tutorial

    Comfy Org · Survey & Overview

    Official tutorial for using Qwen-Image-2512 in ComfyUI workflows for text-to-image generation.

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.