KAT-Coder-Pro V2 is the flagship AI programming model launched by the KwaiKAT team at Kuaishou, deeply optimized for agentic coding scenarios. The model is natively compatible with over 10 mainstream AI coding schemes, including Claude Code, Cline, and OpenClaw...
Matrix-Game 3.0 is a real-time interactive world model developed by Skywork AI, supporting real-time generation at 720p@40FPS. The model employs an error buffering mechanism for self-correction, combined with camera-based perception and memory retrieval to ensure...
Star Office UI is an open-source, pixel-style AI-powered visual status dashboard that supports real-time visualization of the AI assistant's work status. Star Office UI supports mapping six statuses to different office areas, and includes yesterday's work status...
AnyClaw is an open-source smart adapter designed specifically for AI agents, acting as a connector between the Agent Internet and the traditional Internet. It can integrate scattered tools from the traditional Internet, such as APIs, scripts, and web automation workflows...
Gemini 3.1 Flash Live is Google's latest high-quality real-time speech model, designed for natural and fluent conversational interaction. The model has significant improvements in intonation understanding, reasoning ability, and response speed, and can accurately recognize...
Voxtral TTS is an open-source text-to-speech model from Mistral AI, based on a 4 billion parameter architecture and supporting 9 languages. The model boasts an ultra-low latency of 90 milliseconds and a 6x real-time generation speed, capable of generating text from just 3-5 seconds of audio...
MOCR (Multimodal OCR) is a multimodal document parsing model jointly developed by Huazhong University of Science and Technology and Xiaohongshu hi lab. It achieves breakthrough performance in document parsing and image reconstruction with only 3B parameters.
daVinci-MagiHuman is an open-source audio and video co-generation foundation model jointly developed by the GAIR Lab at Shanghai Institute of Innovation and Technology and Sand.ai. The model employs a single-stream Transformer architecture with 15 billion parameters, uniformly modeling text,...
DataChef is an open-source AI data recipe generation model jointly developed by the Shanghai Artificial Intelligence Laboratory and Fudan University. The model automatically generates a complete data processing pipeline for large-scale model adaptation tasks through reinforcement learning, including data selection, cleaning, etc.
Lyria 3 Pro is an AI music generation model launched by Google that supports the creation of complete audio tracks up to 3 minutes long. It has structure awareness capabilities and can accurately process arrangement elements such as intros, verses, choruses, and bridges.
TurboQuant is a vector quantization algorithm developed by Google Research that can compress large model KV cache from 32-bit to 3-bit, achieving a 6x reduction in memory usage, an 8x increase in inference speed, and zero loss of accuracy.
OpenCLI is an open-source command-line tool that transforms any website and Electron desktop application into a command-line interface. The tool connects to a local daemon via a Chrome browser extension, reusing the user's logged-in account permissions...
MiniMax Office Skills is an open-source, production-grade office document engine from Xiyu Technology. The tool comprises four core components: MiniMax-docx (Word), MiniMax-xlsx (Excel), MiniMax-pdf (PDF), ...
Sub2API is an open-source AI API gateway platform that supports unified access and management of subscriptions to mainstream AI services such as Claude, OpenAI, Gemini, and Antigravity. The platform provides multi-account management, API Key distribution, ...
DeerFlow 2.0 is an open-source super agent framework from ByteDance. It employs a 'main agent + 11-layer middleware chain + dynamic sub-agents' architecture, achieving multi-agent collaboration through LangGraph. The framework includes built-in deep research, data analysis, and more...
MAI-Image-2 is the second-generation text-based image model developed by Microsoft's AI Superintelligence team, currently ranking among the top three globally on the Arena.ai leaderboard. The model focuses on three core capabilities: enhanced photorealism,...
New API is a next-generation AI gateway and asset management system. As an AI foundation platform, it provides unified infrastructure access to 30+ mainstream AI services worldwide (OpenAI, Claude, Gemini, DeepSeek, etc.).
PrismAudio is a video-to-audio framework developed by Alibaba's Tongyi Lab, which automatically adds ambient sound effects to silent videos. The model pioneers a 'decomposition-based thought chain' technology, allowing the model to first consider the sound content...
ChatClaw is an open-source AI agent developed by Zhima, with a tiny 30MB installation package supporting macOS/Windows/Linux. The tool includes a built-in local knowledge base, skills marketplace, MCP extensions, long-term memory, and scheduled tasks, and can be integrated with DingTalk, Lark, etc.
Uni-1 is a unified image understanding and generation model from Luma AI, which is the first to integrate visual reasoning and image generation into a single autoregressive Transformer architecture. The model can perform structured internal processing before and during generation...
SoulX-LiveAct is an open-source real-time digital human generation framework from the Soul App AI team, solving the stability challenge of streaming AR diffusion models. Core innovations include: Neighbor Forcing technology to align the diffusion steps between adjacent frames...
gstack is an AI programming workflow for Claude Code, open-sourced by YC CEO Garry Tan, that transforms AI assistants into virtual engineering teams. The tool includes 15 expert roles (such as CEO reviewer, Staff engineer, etc.).
IndexCache is a sparse attention acceleration technology developed by Tsinghua University and Zhipu team. It addresses the problem of high indexer computational overhead in DeepSeek Sparse Attention (DSA) by reducing redundant computation through cross-layer index reuse.
Qwen3.5-Max-Preview is a flagship large-scale model preview version launched by the Ali Tongyi Qianwen team. The model ranked among the top five globally and first in China with a score of 1464 on the LMARaena evaluation platform, and ranked sixth globally in basic ability evaluation.