Quanta AI Labs

03 / Jan 12 – Jan 18, 2026 · 36 RELEASES

AI News Week 3, 2026

Speech AI dominates with Pocket-TTS, VibeVoice-7B, and PersonaPlex-7B. ShowUI-Aloha/π advance GUI agents, Google launches Universal Commerce Protocol, and GLM-Image brings hybrid image generation.

Language Models (2)

  • GLM-4.7-REAP-218B-A32B

    Cerebras · Reasoning

    218B parameter MoE model with 32B active params, vLLM compatible, REAP methodology for enhanced reasoning.

  • TranslateGemma

    Google · Multilingual

    Open translation model supporting 55 languages, 12B outperforms Gemma 3 27B baseline with multimodal capabilities.

Vision & Image (3)

  • GLM-Image

    Z.ai · Text-to-Image

    Hybrid autoregressive + diffusion image model with text-to-image, editing, style transfer, and identity-preserving generation.

  • Qwen-Image-2512 GGUF

    Unsloth · Text-to-Image

    Top open-source diffusion model with realistic people, rich textures, and accurate text rendering via ComfyUI.

  • VIBE

    AI Forever · Image Editing

    Visual instruction-based image editor, powerful open-source framework for text-guided editing.

Video & Animation (2)

  • VerseCrafter

    Research · Text-to-Video

    4D-aware video world model with unified control over camera and multi-object motion via GeoAdapter.

  • PixVerse V5

    PixVerse · Text-to-Video

    AI video generation platform with latest V5 model for high-quality video creation.

Audio & Speech (7)

  • Soprano-1.1-80M

    ekwek · Text-to-Speech

    Ultra-compact 80M TTS model with <1GB memory, 32kHz crystal clear audio, WebUI and OpenAI-compatible endpoint.

  • Supertonic-2

    Supertone · Text-to-Speech

    66M param multilingual TTS, 167× faster than real-time, optimized for on-device deployment.

  • PersonaPlex-7B

    NVIDIA · Voice Agents

    Full-duplex speech model with consistent persona, based on Moshi architecture for natural conversations.

  • VibeVoice-7B

    Microsoft · Text-to-Speech

    Frontier open-source TTS for podcasts and multi-speaker audio with 7.5Hz continuous speech tokenizers.

  • Pocket-TTS

    Kyutai Labs · Text-to-Speech

    Lightweight CPU-only TTS that fits in your pocket, pip install and go with 1.6B delayed streams model.

  • Silero Models

    Silero · Text-to-Speech

    Pre-trained enterprise-grade STT/TTS models with multi-language support via PyTorch hub.

  • NovaSR

    ysharma3501 · Audio Processing

    Tiny 52KB audio upsampler, 16kHz→48kHz at 3500x realtime for TTS enhancement.

3D & Spatial (4)

  • AnyDepth

    AIGeeksGroup · Depth Estimation

    Simple and efficient zero-shot monocular depth estimation with reduced parameters and computational cost.

  • ShapeR

    Meta · 3D Generation

    Robust conditional 3D shape generation from casual captures using Aria glasses pipeline.

  • RigMo

    Research · 3D Generation

    Unified rig and motion learning from mesh sequences with Gaussian bones and skinning weights.

  • UniSH

    HKUST · 3D Reconstruction

    Unified scene and human reconstruction in a single feed-forward pass.

Agents & Automation (5)

  • ToolOrchestra

    NVIDIA · Agent Orchestration

    Small orchestrators managing models and tools for complex reasoning on Humanity's Last Exam benchmark.

  • ShowUI-Aloha

    ShowLab · Desktop Agents

    Human-taught GUI agent that learns workflows from screen recordings with recorder, learner, planner, and actor.

  • ShowUI-π

    ShowLab · Desktop Agents

    450M flow-based VLA model for continuous GUI actions, generating smooth clicks and drags from screen observations.

  • Vision-Agents

    GetStream · Agent Framework

    Open framework for real-time video AI with Stream's edge network for ultra-low latency.

  • Universal Commerce Protocol

    Google · Agent Orchestration

    Open protocol for agentic commerce, co-developed with Shopify, Stripe, Walmart, and 20+ partners.

Developer Tools (9)

  • OpenEnv

    Meta PyTorch · AI Infrastructure

    Interface library for RL post-training with environments including Echo, Code Sandbox, and Oumi integration.

  • Tiny-GPU

    Adam Maj · AI Infrastructure

    Minimal GPU design in Verilog to learn GPU architecture from ground up, 10k+ stars educational project.

  • MCP-CLI

    Phil Schmid · AI Infrastructure

    Lightweight CLI for MCP servers with JSON output, agent-optimized for Gemini CLI and Claude Code.

  • Claude-Mem

    thedotmack · AI Infrastructure

    Claude Code plugin for persistent memory across sessions, captures tool usage and injects context.

  • MLOps Crash Course

    DailyDoseOfDS · AI Infrastructure

    Comprehensive MLOps + LLMOps tutorial series covering foundations to production deployment.

  • Perplexity Prompts Guide

    labeldekho · Prompt Engineering

    Battle-tested prompting strategies for RAG-based search engines like Perplexity AI.

  • pikepdf

    pikepdf · Data Integration

    Python library for reading and writing PDFs, powered by QPDF with Pythonic interface.

  • Instagram-CLI

    supreme-gg-gg · AI Infrastructure

    Terminal UI for Instagram in TypeScript/Python, the ultimate weapon against brainrot.

  • ComfyUI Installers

    ComfyUI Studio · AI Infrastructure

    1-click installers with UV for 100x faster installs, Torch 2.9, CUDA 13, FaceID, and IP-Adapter.

Research & Safety (4)

  • Verbalized Sampling

    Stanford · Research Paper

    Training-free prompting strategy to mitigate mode collapse in LLMs, achieving 2-3x diversity improvement.

  • HeartMula

    Research · Research Paper

    Heart-related medical AI visualization and reconstruction research.

  • SLM Agents Research

    NVIDIA Research · Research Paper

    Research on small language models as the future of agentic AI for specialized applications.

  • Awesome Video Production

    ad-si · Survey & Overview

    Curated list of video production tools, AI generators, teleprompters, and editing software.

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