Quanta AI Labs

52 / Dec 22 – Dec 28, 2025 · 35 RELEASES

AI News Week 52, 2025

GLM-4.7 and MiniMax M2.1 advance coding, Qwen-Image-2512 and HY-Motion-1.0 push generation, with Titans+MIRAS and Stable Video Infinity leading research.

Language Models (6)

  • GLM-4.7

    Zhipu AI · Code Generation

    Advanced coding model with +5.8% on SWE-bench, multilingual agentic coding, terminal tasks, and interleaved thinking for complex reasoning.

  • MiniMax M2.1

    MiniMax · Code Generation

    Enhanced multi-language programming for real-world tasks with native Android/iOS development, vibe coding, and agent scaffolding support.

  • QwenLong-L1.5-30B-A3B

    Tongyi-Zhiwen · Reasoning

    Post-training recipe for long-context reasoning and memory management with 30B parameters and 3B active.

  • LFM2-2.6B-Exp

    Liquid AI · Small Language Models

    Hybrid model designed for edge AI and on-device deployment with transformers, vLLM, and llama.cpp support.

  • Noyron

    LEAP 71 · Reasoning

    Foundational Large Computational Engineering Model encoding physics, manufacturing rules, and domain knowledge.

  • WeDLM-8B-Instruct

    Tencent · Small Language Models

    High-performance 8B instruction-tuned model with fast inference speed and multi-turn conversation support.

Vision & Image (5)

  • Molmo2-8B

    Allen AI · Visual Understanding

    Open vision-language model with video understanding, pointing QA, object tracking, and multi-image capabilities.

  • Generative Refocusing

    NYCU · Image Editing

    Flexible defocus control from a single image with aperture adjustment and interactive refocusing capabilities.

  • Qwen-Image-Edit-2511

    Qwen · Image Editing

    Advanced image editing model with instruction-following capabilities for precise visual modifications.

  • Air Canvas

    Community · Visual Understanding

    Computer vision project for drawing in the air using hand gesture recognition with OpenCV.

  • Qwen-Image-2512

    Qwen · Text-to-Image

    December update of Qwen-Image with enhanced human realism, finer natural detail, and improved text rendering.

Video & Animation (8)

  • FlashPortrait

    Research · Real-time Video

    6x faster infinite-length portrait animation with adaptive latent prediction and ID-preserving video diffusion.

  • InfCam

    Research · Text-to-Video

    Infinite-homography conditioning for robust camera-controlled video generation with improved spatial consistency.

  • StoryMem

    Research · Text-to-Video

    Multi-shot long video storytelling with memory-based keyframe selection for minute-long coherent narratives.

  • ReCo

    Research · Video Editing

    Region-constraint in-context generation for instructional video editing with replacement, addition, and stylization.

  • Spatia

    Microsoft Research · Text-to-Video

    Video generation with updatable spatial memory using 3D scene point clouds for long-horizon consistency.

  • DreaMontage

    ByteDance · Text-to-Video

    Arbitrary frame-guided one-shot video generation for flexible montage creation and video customization.

  • AniX

    Research · Text-to-Video

    Animate any character in any world with action control, character/scene customization, and long-horizon generation.

  • Stable Video Infinity

    EPFL VITA · Text-to-Video

    Generate any-length videos with temporal consistency, scene transitions, and controllable storylines.

3D & Spatial (4)

  • MVInverse

    Research · 3D Reconstruction

    Feed-forward multi-view inverse rendering in seconds with consistent intrinsic image recovery.

  • 3D-RE-GEN

    Research · 3D Reconstruction

    3D reconstruction of indoor scenes with generative framework for separable ground-aligned scene generation.

  • CARI4D

    NVIDIA · 3D Reconstruction

    Category-agnostic 4D reconstruction of human-object interaction from monocular RGB videos with zero-shot generalization.

  • HY-Motion 1.0

    Tencent Hunyuan · 3D Generation

    Billion-parameter DiT model for text-to-3D human motion generation with flow matching.

Agents & Automation (5)

  • Agent Skills

    Anthropic · Agent Framework

    Framework for building modular AI agent capabilities with practical skills, security considerations, and real-world task handling.

  • NitroGen

    NVIDIA MineDojo · Agent Framework

    Foundation model for generalist gaming agents trained on 40k hours across 1000+ games with 52% transfer improvement.

  • Gemini Quant

    lopushok9 · Agent Framework

    Quantitative trading framework powered by Google Gemini for AI-driven financial analysis.

  • SearchAgent-8B

    aidenjhwu · Web Agents

    8B parameter agent model optimized for high-quality web search and information retrieval tasks.

  • Agent Zero

    agent0ai · Agent Framework

    Open-source autonomous AI agent framework for building self-improving agents with memory and tool use.

Developer Tools (6)

  • LYRN

    bsides230 · AI Infrastructure

    Learning framework and repository for AI experimentation and research.

  • TimescaleDB

    Timescale · AI Infrastructure

    Open-source time-series database for fast analytics, real-time AI applications, and scalable data infrastructure.

  • GPU Hot

    Community · AI Infrastructure

    Real-time NVIDIA GPU dashboard with HTTP and WebSocket APIs for monitoring GPU metrics.

  • TerraTorch

    TerraStack AI · AI Infrastructure

    Python toolkit for fine-tuning Geospatial Foundation Models (GFMs) with multi-source data support.

  • Vibe Kanban

    Bloop AI · AI Infrastructure

    Get 10X more out of Claude Code, Codex, or any coding agent with intelligent task management.

  • Late

    Late.dev · AI Infrastructure

    Social media scheduling and analytics API supporting 12 platforms including Instagram, TikTok, and YouTube.

Research & Safety (1)

  • Titans + MIRAS

    Google Research · Research Paper

    Neural architectures for AI long-term memory with surprise-based learning and deep memory modules.

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