Quanta AI Labs

15 / Apr 6 – Apr 12, 2026 · 23 RELEASES

AI News Week 15, 2026

GLM-5.1 tops SWE-Bench Pro as the first open-source model beating GPT-5.4 and Claude Opus. Meta launches Muse Spark multimodal reasoning model with multi-agent orchestration. Anthropic announces Project Glasswing for AI-powered cybersecurity. LPM 1.0 achieves first full-duplex conversational video generation. Waypoint 1.5 and InSpatio-World advance real-time 3D worlds. ACE-Step 1.5 brings commercial-grade music generation to consumer GPUs.

Language Models (2)

  • GLM-5.1

    Zhipu AI (Z.ai) · MoE Architecture

    Open-source 744B MoE flagship model that tops SWE-Bench Pro at 58.4, beating GPT-5.4 and Claude Opus 4.6. Designed for agentic engineering with 8+ hour autonomous task execution, released under MIT license.

  • Muse Spark

    Meta Superintelligence Labs · Reasoning

    Natively multimodal reasoning model with tool-use, visual chain of thought, and multi-agent orchestration via Contemplating Mode that enables parallel reasoning across multiple agents.

Vision & Image (3)

  • VANAST

    Seoul National University · Image Editing

    Photorealistic virtual try-on system that transfers clothing onto animated human images conditioned on pose video while preserving identity. CVPR 2026 highlight paper.

  • Anima

    CircleStone Labs / Comfy Org · Text-to-Image

    2B-parameter text-to-image model specialized in anime and non-photorealistic art generation, built on NVIDIA Cosmos and trained on millions of anime images plus 800K artistic images.

  • Open Generative AI

    Anil-matcha · Text-to-Image

    Free open-source alternative to Higgsfield AI, Freepik, and Krea, providing text-to-image, image-to-image, video generation, and lip-sync using 200+ AI models via MuAPI.

Video & Animation (3)

  • LPM 1.0

    Anuttacon (miHoYo) · Real-time Video

    17B-parameter diffusion transformer for real-time full-duplex conversational video generation. First model to achieve simultaneous speaking, singing, listening with unlimited-length output across photorealistic, anime, and 3D character styles.

  • NUMINA

    Huazhong University / Zhejiang University · Text-to-Video

    Training-free framework that fixes incorrect object counts in text-to-video models by analyzing attention maps and using layout-guided attention modulation to ensure accurate object quantities.

  • MMPhysVideo

    CASIA / StepFun · Text-to-Video

    Framework improving video generation physical accuracy through joint multimodal modeling, combining semantic, geometric, and trajectory information for physically realistic and visually consistent content.

Audio & Speech (1)

  • ACE-Step 1.5

    StepFun · Music Generation

    Open-source music generation model producing commercial-grade audio from text prompts on consumer GPUs. Supports 10s to 10min durations with cover generation, vocal-to-BGM conversion, and a diffusion transformer architecture.

3D & Spatial (4)

  • Waypoint 1.5

    Overworld · 3D Generation

    Real-time AI world simulation engine running at 720p/60fps on consumer GPUs. Trained on 100x more data than v1, supports Mac and Windows with browser-based streaming option.

  • InSpatio-World

    InSpatio · 3D Reconstruction

    Transforms videos into interactive 4D environments with free-viewpoint exploration and temporal control, maintaining persistent world state anchored to reference footage with physical realism.

  • Kimodo

    NVIDIA · 3D Generation

    Kinematic motion diffusion model trained on large-scale optical mocap data, generating high-quality 3D human and robot motions controlled through text prompts and kinematic constraints like poses and joint positions.

  • FML (Floorplan Markup Language)

    University of Tokyo · 3D Generation

    Represents structural floorplan information as markup-based sentences, transforming vector floorplan generation into a next-token prediction task. CVPR 2026 paper.

Agents & Automation (3)

  • MemPalace

    Milla Jovovich / Ben Sigman · Agent Framework

    Open-source AI memory system that stores conversations locally in a structured palace architecture (wings, rooms, halls), achieving 96.6% recall on LongMemEval benchmarks without cloud services.

  • Rowboat

    Rowboat Labs (YC S24) · Agent Framework

    Open-source local-first AI assistant that builds a knowledge graph from email and meeting notes, then uses that context for drafting, meeting prep, and document generation with BYO LLM support.

  • GBrain

    Garry Tan · Agent Framework

    Personal knowledge management system that creates a searchable AI-augmented brain from meetings, emails, and notes, enabling AI agents to maintain long-term memory via markdown-based knowledge base.

Developer Tools (5)

  • Project Glasswing

    Anthropic · AI Infrastructure

    AI-powered cybersecurity initiative using Claude Mythos Preview to identify and fix software vulnerabilities in critical infrastructure, partnered with AWS, Apple, Google, Microsoft, NVIDIA, and others.

  • RotorQuant

    Scrya · AI Infrastructure

    KV cache compression for LLMs using block-diagonal rotations instead of full-rank transforms, achieving better perplexity, faster decode speeds, and 44x fewer parameters than Google's TurboQuant.

  • Code Review Graph

    tirth8205 · AI Infrastructure

    MCP-integrated tool that creates structural codebase maps using Tree-sitter, providing precise context to AI coding assistants and reducing token consumption by 8.2x during code review.

  • Fli

    Punit Arani · AI Infrastructure

    Python MCP server and CLI for programmatic Google Flights search, enabling AI assistants like Claude to search flights through natural language conversation.

  • MuAPI

    MuAPI / Vadoo · AI Infrastructure

    Comprehensive API platform providing unified access to numerous AI models for generating and editing images, videos, and audio content through a single interface.

Research & Safety (2)

  • Video-RAG

    NeurIPS 2025 · Research Paper

    Enhances long video understanding by integrating RAG with large vision-language models, retrieving visually-aligned auxiliary text data including OCR, speech transcription, and object detection.

  • SpatialEdit

    Research Team · Benchmark

    Benchmark and toolkit for evaluating fine-grained image spatial editing including object motion, rotation, 3D viewpoint changes, framing, and camera movement while maintaining visual quality.

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