MAI-Transcribe-1 is an enterprise-grade speech-to-text model launched by Microsoft Azure AI Foundry. It supports 25 languages including Chinese, English, Japanese, and French, and the model outperforms Whisper-large-v3 in the FLEURS benchmark test.
Wan2.7-Video is an AI video creation model developed by Alibaba's Tongyi Lab, supporting full-modal input including text, images, video, and audio. The model breaks through traditional generation limitations, enabling local editing of videos like photo editing...
Gemma 4 is Google's open-source multimodal large model series, based on the Gemini 3 architecture. Models include E2B/E4B (runs on mobile phones/Raspberry Pi), 26B MoE (only 3.8B parameters are activated), and 31B Dense (Arena-rated open-source model)...
InternSVG is a comprehensive "data-evaluation-model" suite for unified SVG modeling, developed by the Shanghai Artificial Intelligence Laboratory and other institutions. It comprises three main components: the SAGogge dataset with over 16 million samples, the standardized SArena benchmark, and...
JoyAI-Image-Edit is an open-source, command-guided image editing model from JD.com, built upon the JoyAI-Image multimodal base model. The model supports precise and controllable spatial manipulation and intelligent image modification via natural language commands...
GLM-5V-Turbo is a native multimodal coding foundation model launched by Zhipu AI, designed specifically for visual programming and AI agents. The model deeply integrates visual and textual capabilities from the pre-training stage, supporting the understanding of images, videos, design drafts, etc.
Qwen3.6-Plus is an Agent programming model developed by Alibaba's Tongyi Lab, focusing on significantly enhancing the capabilities of encoded intelligent agents. The model supports ultra-long contexts with up to 1 million tokens by default, and includes code fixes in SWE-bench, Terminal-Bench, and other benchmarks...
Claw-code is a Clean-room Rewrite project launched after the Claude Code source code was leaked. To avoid legal risks, the authors used OpenAI Codex's OmX workflow orchestration, refactoring the core architecture from Typ...
Wan2.7-Image is an AI image generation and editing model developed by Tongyi Labs. It supports deep customization of character images (face shape, eye shape, bone structure, etc.), bidding farewell to the monotonous "AI standard face," and can accurately generate 4000+ characters and...
Lightpanda is an open-source headless browser designed specifically for AI agents and automation. Written from scratch in the Zig language, it does not rely on Chromium or WebKit. The tool's memory footprint is only 1/16th that of Chrome, and its execution speed...
Veo 3.1 Lite is Google's latest AI video generation model, costing over 50% less than Veo 3.1 Fast while maintaining the same generation speed. The model supports video generation from text and images and allows flexible switching between portrait and landscape orientations...
TeamClaw is an AI employee team management platform, positioned as a 'one-person AI office'. The platform allows users to define AI roles (CEO, analyst, etc.) just like hiring real employees, with roles being assigned instantly; it supports installing...
ColaMD is an open-source Markdown editor. The tool solves the pain point of traditional editors where manual refreshing is required to see updates after AI modifications, achieving real-time synchronization between agent-based background changes and foreground content, as if viewing...
Harrier-OSS-V1 is an open-source multilingual text embedding model from Microsoft, achieving state-of-the-art (SOTA) performance on the Multilingual MTEB v2 benchmark. The model employs a decoder-only architecture, utilizing final token pooling and L2 regression...
LongCat-AudioDiT is a high-fidelity diffusion-based text-to-speech (TTS) model open-sourced by Meituan. Its core innovation lies in directly generating the text through diffusion in the waveform latent space, rather than using intermediate representations like Mel-frequency spectra, effectively avoiding error accumulation...
Vision2Web is a visual website development evaluation benchmark jointly launched by Tsinghua University and Zhipu AI. It is used to evaluate the end-to-end website building capabilities of multimodal AI agents. Vision2Web includes 193 real-world website tasks, divided into three progressively more difficult levels...
wecom-cli (Enterprise WeChat CLI) is an open-source command-line tool from the official Enterprise WeChat team, enabling humans and AI agents to operate Enterprise WeChat from the terminal. The tool covers contacts, to-do lists, meetings, messages, calendars, documents, etc.
PixVerse V6 is the latest AI video generation model launched by PixVerse Technology. The model achieves breakthroughs in camera control, character performance, and multi-camera audio-visual generation, resulting in more precise camera movements, smoother cross-frame character emotions, and more refined physical interactions...
Pretext is an open-source, pure JavaScript/TypeScript multiline text measurement and layout library created by Cheng Lou, a former React core member. It is used to accurately measure the size of multiline text using pure arithmetic methods, completely bypassing DOM reflow.
Pascal Editor is a web-based, open-source 3D architecture editor developed with the assistance of Claude. The tool requires no installation, runs in a browser, and utilizes WebGPU technology for high-performance rendering.
Qwen3.5-Omni is a full-modal large-scale model launched by Alibaba's Tongyi Lab, capable of simultaneously understanding text, images, audio, and audio-video input. The model employs a Thinker-Talker collaborative architecture and Hybrid-MoE technology, handling 215 audio...
Mureka V9 is Kunlun Wanwei's latest AI music model. The model surpasses competitors like Suno V5 in core metrics such as musical melody, expressiveness, arrangement, and composition, topping the charts in Artificial Analysis for vocals and instruments...
GigaWorld-1, developed by Excellent Vision, is an embodied world model that topped the WorldArena rankings. The model utilizes the AC-WM (Motion Controlled World Model) architecture, combining explicit motion modeling with differentiable physics...
MicroCoder is a code-based large-scale model training optimization project launched by Microsoft Research Asia in collaboration with the University of Cambridge and Princeton University. It addresses the training bottlenecks of next-generation models such as Qwen3 by comprehensively optimizing training across four dimensions: algorithm, data, evaluation, and experience...