Gemini 3.5 Transcribe is Google's latest speech-to-text model, supporting both real-time streaming and pre-recorded audio processing modes. The former achieves sub-second latency through the Live API, while the latter supports speaker attribution...
Omarchy is a Linux distribution created by DHH, the author of Ruby on Rails. It's based on Arch Linux and the Hyprland tiled window manager. Omarchy's core concept is a "flexible computer," where the desktop adapts to the user's...
Qwen3.8-Flash is a multimodal MoE model launched by Alibaba Cloud Tongyi Qianwen. It has a total of 125B parameters, with only 6B activated per token, and natively supports 262K contexts.
GLM-5.3-Flash is the latest open-source large-scale model released by Zhipu, with a total of 32 billion parameters and only 1.8 billion activation parameters. It is the first native multimodal model in the GLM-5 series. The model's overall intelligence index reaches 57 points, comparable to Claud...
QuickDesk is the world's first open-source AI-native remote desktop with a built-in MCP Server, built on Google Chromium Remoting. The tool allows AI agents such as Claude and Cursor to directly take over any remote...
Breeze TTS 2 is a new generation of voice synthesis model launched by BreezeBlue. It supports zero-sample design of character timbre through natural language and precise control of performance details such as emotion, speech rate, and accent using natural language commands.
PixVerse R2 is a real-time, multimodal world model developed by Aishike Technology, an upgraded version of PixVerse R1. The model supports multimodal input including text, images, audio, and motion signals, and can continuously receive user control signals during operation...
EgoSuite-Open100K is the world's first open-source dataset of 100,000 hours of full-modal human behavior, launched by Lightwheel Intelligence, targeting physical AI and embodied intelligence. The dataset contains over 15,000 first-person videos of real-world scenarios, covering various aspects...
OpenStory is an open-source AI video production platform focused on converting scripts into short, stylistically consistent videos with a single click. The platform can automatically break down scenes, design shots, analyze emotions, and utilize mature models like Fal.ai to generate graphics...
Agnes 2.5 Pro Alpha is an open-source multimodal inference model from Agnes AI, licensed under the Apache 2.0 license, with weights hosted on Hugging Face. The model supports text and image input, and its context window can hold up to 1 million tokens...
Huashu-excel is an open-source AI data analysis skill developed by Huashu. It solves the core problem of preventing AI from making calculation errors and reporting them through an eight-step process, including physical examination, cleaning, alignment, analysis, reconciliation, delivery, image verification, and quality control.
ScienceClaw is a research-native intelligent agent platform launched by Zidong Taichu (incubated by the Institute of Automation, Chinese Academy of Sciences). It relies on Zidong Taichu's multimodal large model and covers fields such as life sciences, materials, chemistry, physics, and astronomy.
Faraday is an AI scientist agent developed by the London-based AI lab Inherent. It's based on the Qwen 3.6 model with 27 billion parameters and trained through reinforcement learning. Faraday can perform tasks like a human PhD student without prior knowledge...
OpenBot is an open-source enterprise-grade AI agent platform based on CopilotKit, positioned as a 'digital colleague you can entrust with real-world tasks'. Each agent is equipped with its own dedicated 'computer'. It's an independent Docker container containing a Chromium browser...
TabTin is a full-stack, open-source AI agent team collaboration platform. The platform truly integrates agents into team workflows; tasks can be initiated on desktops, viewed on mobile devices, executed by server agents, and can be handed over with complete context...
FreeToken is an open-source edge-side MoE (Model-Based Inference) system for large models, jointly developed by UC Berkeley, MIT, and other institutions. The system utilizes full-layer double buffering, bandwidth-adaptive hybrid scheduling, agent state reuse, and elastic memory hot-scaling/scaling, among other features...
CoinVE-200K is a large-scale combinatorial instruction video editing dataset open-sourced by Tencent Video's Intelligent Creation Team. It contains 200,000 pairs of 1080p video editing samples, each containing 2-5 atomic editing instructions, including addition, removal, modification, etc.
LLM-as-a-Verifier is a general-purpose verification framework jointly open-sourced by Stanford, UC Berkeley, and NVIDIA Research. The framework requires no additional training and generates continuously fine-grained verification by using the complete logits distribution of LLM-scored tokens...
SenseNova U1.5 Lite is a lightweight, native unified multimodal large model open-sourced by SenseTime, designed specifically for real-world visual creation workflows. The model natively supports 3-4k character ultra-long instructions and features high-quality visual generation...
DeepSeek-V4-Flash-Vision-Exp is an experimental multimodal visual understanding model from DeepSeek. While retaining the text input capabilities of DeepSeek-V4-Flash, it adds visual input, and a multimodal agent...
Cocode is an open-source distribution of DeepSeek Harness, providing both a desktop GUI and a terminal TUI. The tool integrates files, Git, terminal, browser, and agent into a single workspace, with sessions synchronized in real-time. Agen...
Qwen-UI-Agent is a GUI intelligent agent foundation model developed by Alibaba's Tongyi team. It enables users to interact with interfaces on mobile phones, computers, and web browsers much like a real person. The model was trained on 100+ real devices and 150+ applications, and supports GUI...
Lumos NexCore, launched by Lumos Robotics, is an embodied intelligence evolution engine positioned as a skills infrastructure for industry applications. Lumos NexCore integrates data assets, model training, evaluation and verification, skills management, and robot...
Qwen3.8-27B-DFlash2 is an open-source speculative decoding draft model from Inco AI, specifically designed for Qwen3.8-27B. The model has only 1.92B parameters and achieves this through block-level parallel prediction, a lightweight path selector, and dual-tap dynamic convolution...