DiffusionGemma is an experimental open-source text diffusion model released by Google DeepMind. The model is built upon the Gemma 4 architecture and Gemini Diffusion research, employing a 26B MoE design and utilizing parallel diffusion...
Khala is an open-source AI music model jointly developed by the Central Conservatory of Music and Tsinghua University, designed for high-fidelity song generation. The model employs a 64-layer deep acoustic token hierarchy and supports generating music based on text descriptions and lyrics...
North Mini Code is an open-source agentic coding model from Cohere. The model uses the MoE architecture, has a total of 30 bytes of parameters, only 3 bytes of activation, and supports 256K context and 64K output.
MusaCoder is an open-source code model developed by Moore Threads for generating high-performance CUDA/MUSA kernels from PyTorch operators.
Nextie Alpha is the industry's first cognitive model, launched by Nextie, requiring only 4 parameters for edge deployment. Led by Li Di, the "father of Xiaoice," the model utilizes reinforcement learning to decouple knowledge and cognition based on an open-source inference model, condensing...
cmux is an open-source terminal tool designed specifically for AI Coding Agents. It aggregates and displays Git branches, PR status, ports, and Agent notifications through vertical tabs, and uses an intelligent notification ring to accurately locate tasks awaiting confirmation.
Claude Fable 5 is Anthropic's most powerful AI model, the flagship thos-class safe model. Its underlying architecture is completely identical to Claude Mythos 5, but with added conservative safety features.
SCAIL-2 is the second-generation film-grade character animation framework open-sourced by Zhipu in collaboration with Professor Liu Yongjin's research group at Tsinghua University. Based on the Diffusion Transformer (DiT) architecture, it directly achieves high-quality animation transfer through end-to-end training, eliminating the need for traditional...
Gemini 3.5 Live Translate is Google's latest real-time translation model, supporting near real-time speech-to-speech translation for 70+ languages.
Bigset is an open-source multi-agent system developed by the American startup TinyFish, licensed under the AGPL-3.0 license. Users simply describe the required data in natural language, and the system can automatically infer table structures, dispatch agents from real-time networks, etc.
WBench, launched by Meituan's LongCat team, is the first systematic multi-round evaluation benchmark for interactive video world models. It includes 289 test cases and 1058 interaction rounds, covering six themes such as nature, city, and fantasy...
VitaBench 2.0 is the first benchmark for evaluating intelligent agents based on long-term dynamic user modeling in real-life scenarios, launched by Meituan's LongCat team. It includes 56 simulated users, 819 complex tasks, over 2000 dynamic preferences, and 6...
ABot-Earth 0.5 is the world's first native 3D city world model launched by Gaode Maps, a subsidiary of Alibaba Group, positioned as an automated 3D city factory. Users input a single satellite image or text description, and the system can run on a consumer-grade GPU for up to 10...
Kairos-HomeWorld is the world's first fully interactive 3D world model of a house, developed by Daxiao Robotics in collaboration with the Chinese University of Hong Kong and Shenzhen Hetao College. The model utilizes a four-stage hierarchical generation architecture to achieve seamless transitions from text to structure...
CopilotKit is an open-source AI-native application front-end operating system. The tool supports three generative UI modes: controlled, declarative A2UI, and open, allowing AI agents to move beyond chat windows and directly interact with applications...
Open Code Review is an open-source AI code review CLI tool from Alibaba, originating from the group's production-level practices that have served tens of thousands of developers and executed over a million review tasks.
PawBench is a general-purpose intelligent agent evaluation benchmark launched by Tongyi Labs. It is designed for personal assistant and agent scenarios and incorporates the base model and operating framework (Harness) into the same evaluation system.
PlanningBench is an open-source framework launched by Tencent Hunyuan Team in collaboration with Renmin University of China's Gaoling School of Artificial Intelligence and other institutions. It focuses on evaluating and training the planning capabilities of large language models.
GitNexus is an open-source code knowledge graph engine that uses Tree-sitter parsing to build complete dependency graphs from code repositories and exposes these graphs to AI tools such as Cursor, Claude Code, and Codex using the MCP protocol.
Magenta RealTime 2 (MRT2) is the second-generation open-source local real-time music generation model released by Google's Magenta team. MRT2 is based on a frame-by-frame generation architecture, with each frame taking 40 milliseconds, achieving an actual control latency of approximately 200 milliseconds...
html-video is an open-source HTML version of CapCut, developed by the OpenDesign team (nexu.io) and built on the hyperframes framework. The tool allows agents to generate world-class product promotions and explanations by writing HTML...
Gemma 4 12B is an open-source multimodal large model from Google. It adopts the industry's first encoder-free unified architecture, with visual and audio data directly input into the LLM backbone, completely eliminating the need for a separate encoder.
BrowserAct Skills is a browser automation CLI tool for AI Agents that addresses core pain points when agents control browsers, such as cookie-free environments, anti-scraping interception, CAPTCHA blocking, and gaps in human-computer collaboration.
Microsoft Scout is an AI personal assistant launched by Microsoft, built on the OpenClaw open-source technology. The tool has an independent Entra identity and can run continuously and autonomously in the background without requiring users to trigger it each time.