Fun-CosyVoice 3.5 is the latest speech generation model released by the speech team at Alibaba's Tongyi Labs, focusing on multilingual timbre replication and refined expression control. The model's most prominent innovation lies in its FreeStyle natural language control capabilities—...
FireRed-OCR is a lightweight visual language model for document structure parsing, open-sourced by the Xiaohongshu team. With only 2 bytes of parameters, it achieved a 92.94% overall score in the authoritative OmniDocBench v1.5 benchmark, surpassing GPT-5.2, ...
Fun-AudioGen-VD is an innovative large-scale speech model launched by the speech team of Alibaba Tongyi Labs. It is positioned as a professional tool for 'sound design and contextualized audio generation'. The model supports 'FreeStyle' free command generation and can...
Goose is an open-source local AI agent framework from Block that can autonomously execute complete engineering tasks—reading files, writing code, running tests, calling APIs, automatically debugging, and correcting itself after errors until completion.
DataClaw is an open-source AI dialogue data export tool developed by Peter O'Malle. It can automatically capture the complete dialogue history between users and AI programming assistants such as Claude Code, Codex CLI, and Gemini CLI, and convert it into...
OpenSandbox is an open-source, general-purpose AI application sandbox platform from Alibaba, licensed under the Apache 2.0 license. It is designed for securely executing AI-generated code and automating tasks. It provides multi-language SDKs (Python, Java/Ko...).
TinyClaw is a lightweight, open-source multi-agent collaboration framework developed by TinyAGI, designed for resource-constrained environments. TinyClaw supports running multiple specialized AI agents (such as @coder and @writer) simultaneously on devices like Raspberry Pi...
OpenFang is an open-source agent operating system built in Rust. Its core innovation is Hands—seven pre-built packages of autonomous capabilities that allow AI to automatically perform tasks 24/7 without manual intervention (such as monitoring competitors, data mining, etc.).
AngelSlim is a multimodal large model compression toolkit developed and open-sourced by Tencent's Hunyuan team. It utilizes techniques such as quantization, speculative sampling, sparsification, and distillation to compress large language models (LLM), visual language models (VLM), and speech models...
Arrow 1.0 is Quiver AI's first native AI-generated SVG model. Unlike traditional pixel-based generation models, it directly outputs scalable vector graphics code. The images are constructed based on mathematical formulas and support infinite...
pplx-embed is a series of text embedding models launched by Perplexity, including pplx-embed-v1 for standard retrieval and pplx-embed-context-v1 for context awareness, both available in 0.6B and 4B sizes.
SkyReels V4 is a video foundation model developed by Kunlun Tech. It is the world's first AI video model to support multimodal input, joint audio and video generation, and unified generation/repair/editing. The model employs a dual-stream MMDiT architecture and can generate 108...
LocoOperator-4B is an open-source 4B parameterized code exploration agent from LocoreMind, trained from Qwen3-Coder-Next through knowledge distillation based on Qwen3-4B-Instruct. The model is positioned as a Claude Coder...
Nano Banana 2 is a next-generation image generation model (Gemini 3.1 Flash Image) from Google DeepMind. The model integrates the Gemini knowledge base and real-time web search, enabling it to accurately render real-world scenes and generate...
Ctrl-World is an embodied world model jointly developed by Chen Jianyu from Tsinghua University and Chelsea Finn's team from Stanford University. In the authoritative evaluation by WorldArena, it won first place globally in embodied task capability and second place globally in video generation quality.
Talksign-1 is an AI real-time translation model designed specifically for American Sign Language (ASL). It supports bidirectional conversion and can capture 3D human key points through a camera to recognize 250 ASL words, or convert speech/text into sign language video.
IronClaw is an open-source AI assistant developed by the NearAI team. Implemented in Rust, it prioritizes local compatibility and security. IronClaw uses a WASM sandbox for execution and manages credentials through an encrypted vault...
MimiClaw is an ultra-lightweight AI assistant based on the ESP32-S3 chip. It is written in pure C language and can run directly on bare metal without Linux, Node.js or an operating system.
NanoClaw is an open-source AI assistant that serves as a lightweight alternative to OpenClaw. Each agent runs in an independent sandbox and only accesses explicitly mounted directories.
Zvec is a lightweight, embedded vector database open-sourced by Alibaba, built on its internal Proxima engine. Unlike traditional standalone deployment solutions, Zvec runs directly within the application process, requiring no additional servers or configuration; installation is straightforward...
MoChat is an AI-native social platform launched by the Data Science Lab at the University of Hong Kong. Through adapters such as OpenClaw, Nanobot, and Claude Code, the platform allows AI agents to automatically discover potential collaborators with independent identities...
ClawFeed is an open-source AI news summarizing tool that helps users aggregate information sources from multiple platforms such as Twitter, RSS, HackerNews, Reddit, and GitHub, and generates structured summaries through AI intelligent filtering (supporting 4 hours...).
OpenAkita is an open-source, self-evolving AI assistant framework. It can execute commands like a traditional AI assistant, possessing a unique 'self-evolution' capability that automatically performs memory consolidation, error correction, and task review. When it encounters a bottleneck, it can automatically...
FireRedASR2S is an industrial-grade end-to-end speech recognition model open-sourced by Xiaohongshu's Super Intelligence-AudioLab, integrating four state-of-the-art (SOTA) modules: ASR, VAD, language recognition, and punctuation prediction.