Quanta AI Labs

04 / Jan 19 – Jan 25, 2026 · 41 RELEASES

AI News Week 4, 2026

Desktop agents explode with Openwork, Eigent, and Claude Cowork variants. StepFun releases Step3-VL-10B and Step-Audio-R1.1, DeepSeek introduces Engram memory architecture, and xAI open-sources the X algorithm.

Language Models (3)

  • GLM-4.7-Flash

    Z.ai · MoE Architecture

    30B-A3B MoE model, strongest in 30B class with optimal balance of performance and efficiency for lightweight deployment.

  • Engram

    DeepSeek · Reasoning

    Conditional memory via scalable lookup - new sparsity axis for LLMs with U-shaped scaling law and Engram-27B verification.

  • GLM-4.7-Flash GGUF

    Unsloth · Small Language Models

    Quantized GGUF version of GLM-4.7-Flash for efficient local deployment with llama.cpp compatibility.

Vision & Image (4)

  • Step3-VL-10B

    StepFun · Visual Understanding

    10B multimodal model rivaling 10-20× larger models like GLM-4.6V and Gemini 2.5 Pro with parallel coordinated reasoning.

  • MedGemma 1.5-4B

    Google · Visual Understanding

    Medical AI with 3D CT/MRI, histopathology, longitudinal imaging, EHR understanding, and anatomical localization.

  • Medical-SAM3

    AIM Research · Visual Understanding

    Foundation model for universal prompt-driven medical image segmentation across multiple imaging modalities.

  • OmniTransfer

    Research · Image-to-Image

    Universal style and content transfer across modalities with advanced diffusion techniques.

Video & Animation (10)

  • Waypoint-1

    Overworld · Real-time Video

    Real-time interactive video diffusion controllable via text, mouse, and keyboard with zero-latency frame generation.

  • VIGA

    Research · Text-to-Video

    Advanced video generation and animation research with novel visual synthesis techniques.

  • VideoMaMa

    KAIST CVLab · Text-to-Video

    Video generation research with advanced motion modeling and temporal consistency.

  • CoDance

    Lucaria Academy · Text-to-Video

    Coordinated dance generation research for multi-person synchronized motion synthesis.

  • FrankenMotion

    Research · Video Editing

    Motion composition and editing framework for flexible character animation control.

  • FlowAct-R1

    Research · Text-to-Video

    Flow-based action generation with reasoning capabilities for realistic motion synthesis.

  • Motion3-to-4

    Research · Text-to-Video

    3D to 4D motion generation lifting static poses into dynamic temporal sequences.

  • MOTIVE

    NVIDIA · Text-to-Video

    Motion generation research for interactive and realistic character animation.

  • Remotion

    Remotion · Video Editing

    Make videos programmatically with React - compose, edit dynamically, and render at scale.

  • Higgsfield Creators

    Higgsfield AI · Text-to-Video

    AI video creation platform with YouTube campaign tools for content creators.

Audio & Speech (7)

  • VoxCPM1.5

    OpenBMB · Text-to-Speech

    Tokenizer-free TTS with 44.1kHz audio, context-aware generation, true-to-life voice cloning, and LoRA fine-tuning support.

  • Step-Audio-R1.1

    StepFun · Voice Agents

    Realtime speech with dual-brain architecture enabling thinking while speaking, acoustic-grounded reasoning, and ultra-low latency.

  • Chroma-4B

    FlashLabs · Voice Agents

    Multimodal virtual human model with speech understanding, text+speech generation, and voice cloning on Qwen2.5-Omni backbone.

  • LuxTTS

    ysharma3501 · Text-to-Speech

    High-quality TTS with 150× realtime speed, SOTA voice cloning, 48kHz clarity, fits in 1GB VRAM.

  • VibeVoice-ASR

    Microsoft · Speech-to-Text

    Unified ASR handling 60-minute audio in single pass with speaker diarization, timestamps, and customized hotwords.

  • PersonaPlex

    NVIDIA · Voice Agents

    Full-duplex speech model research for persona-consistent natural conversations.

  • ComfyUI FL-HeartMuLa

    filliptm · Music Generation

    Multilingual AI music generation nodes for ComfyUI with 4-minute songs, section markers, and style tags.

Agents & Automation (8)

  • Openwork

    Accomplish AI · Desktop Agents

    Open-source AI desktop agent that acts autonomously - local-first, BYOK, runs on your computer.

  • Dexter

    virattt · Agent Framework

    Autonomous agent for deep financial research with task planning, self-validation, and real-time market data.

  • Claude Cowork for Office

    Cowork Studio · Agent Framework

    Office-focused AI agent with 20+ tools, document writing, code development, GUI/CLI modes, and MCP support.

  • Open Claude Cowork

    Composio · Agent Framework

    Open-source Claude Cowork with 500+ SaaS integrations, multi-provider support, and skills system.

  • Clawdbot

    Clawdbot · Agent Framework

    Personal AI assistant with multi-channel inbox (WhatsApp, Telegram, Slack, Discord, iMessage), voice wake, and live canvas.

  • Reachy Mini Desktop

    Pollen Robotics · Robotics

    Robot control app with real-time 3D visualization, Hugging Face app store, and cross-platform support.

  • Reachy Mini Reactions

    Pollen Robotics · Robotics

    Hugging Face Space demo for Reachy Mini robot reactions and interactions.

  • Loot Drop

    Loot Drop · Agent Framework

    AI-powered platform for gaming and digital asset discovery.

Developer Tools (7)

  • Linum V2

    Linum AI · AI Infrastructure

    Advanced AI-powered field notes and documentation platform launch.

  • X-Algorithm

    xAI · AI Infrastructure

    Open-source algorithm powering X For You feed with Grok-based transformer ranking and Phoenix retrieval.

  • Cognee

    Topoteretes · Data Integration

    Memory for AI agents in 6 lines - replaces RAG with unified graph+vector layer, 30+ data sources.

  • Semantica

    Hawksight AI · Data Integration

    Semantic intelligence framework for GraphRAG, multi-agent systems, and high-stakes enterprise AI.

  • Reasoning from Scratch

    Sebastian Raschka · AI Infrastructure

    Build a reasoning LLM in PyTorch from scratch - companion repo for Manning book on DeepSeek R1-style models.

  • Sweep Next-Edit 1.5B

    Sweep AI · Inference Engine

    Predicts next code edit in <500ms locally, outperforms 4× larger models, GGUF format with JetBrains plugin.

  • BytePlus AI

    BytePlus · AI Infrastructure

    Enterprise AI platform to accelerate AI journey with ByteDance infrastructure.

Research & Safety (2)

  • Awesome NVIDIA

    Yash Kavaiya · Survey & Overview

    Curated list of NVIDIA resources, tools, frameworks, and learning materials for AI developers.

  • RL Video Overview

    YouTube · Survey & Overview

    Educational video overview of reinforcement learning concepts and applications.

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