VerseCrafter is a dynamic, realistic video world model developed by Fudan University, Tencent PCG ARC Lab, and other institutions, possessing 4D geometric control capabilities. The model is trained on the large-scale real-world dataset VerseControl4D...
NovaSR is an open-source audio super-resolution model that, with only 52KB of memory, can upsample low-sampling-rate audio (such as telephone quality) from 16kHz to high-sampling-rate audio (such as studio-quality) from 48kHz. The model uses a neural network pre-processing...
Playwriter is an open-source Chrome extension built on Microsoft Playwright and designed specifically for AI-powered browser operations. The tool seamlessly integrates with Chrome as an extension, directly embedding AI capabilities into the user's browsing experience...
Prompt Manager (PromptX) is a tool for efficiently managing AI prompts, designed to improve AI workflow efficiency. The tool supports the creation, editing, categorization, and version management of personal prompts. Through intelligent categorization and global...
FrogBoss is a series of programming models developed by Microsoft Research. The models are trained using an innovative and sophisticated vulnerability generation method (BugPilot), which involves having a Language Model Agent (LLM) add new functionality to the codebase...
OpenWork is an open-source desktop application that provides knowledge workers with a Claude Cowork-style workflow. Built on OpenCode, OpenWork offers a clean, guided user interface.
TranslateGemma is a new generation of open-source translation models from Google, based on Gemma 3. The model supports high-quality translation of 55 languages through a combination of supervised fine-tuning and reinforcement learning, significantly improving translation quality...
FLUX.2 [klein] is an open-source, high-efficiency image generation and editing model from Black Forest Labs. The model boasts sub-second inference speeds, capable of generating and editing high-quality images within 0.5 seconds, supporting text-to-image and image-to-image processing...
AgentCPM-Explore is an open-source agent model jointly developed by Tsinghua University, Renmin University of China, Wallfacer Intelligence, and the OpenBMB open-source community. Based on only 4B parameters, the model outperforms similar models on multiple long-term task benchmarks...
ArenaRL is an open-source contrastive reinforcement learning method for open-domain agents, developed by the Tongyi DeepResearch team in collaboration with AutoNavi. ArenaRL introduces a tournament mechanism, transforming traditional absolute scoring into relative ranking within groups...
Step-Audio-R1.1 is the world's first open-source native speech inference model, developed by Step-Audio. The model topped the global authoritative speech inference leaderboard with an accuracy of 96.4%, surpassing many leading models. The model features deep speech inference capabilities,...
Sinon (Sinon Large Language Model) is a large-scale vertical model for the agricultural field launched by Nanjing Agricultural University. Leveraging the disciplinary strengths of Nanjing Agricultural University, the model has collected over 4 billion tokens of data covering multiple agricultural disciplines, including books, papers, etc.
OctoCodingBench is an open-source benchmark suite from MiniMax for evaluating the instruction compliance capabilities of Coding Agents. OctoCodingBench simulates real-world software development scenarios, assessing performance based on system constraints, user instructions, project specifications, skills, etc.
MedGemma 1.5 is an open-source, multimodal AI medical model from Google, specifically designed for processing medical images and text data. The model supports high-dimensional medical images (such as CT and MRI), whole-slice pathological images, longitudinal image analysis, anatomical localization, etc.
GLM-Image is a new generation of multimodal image generation model open-sourced by Zhipu in collaboration with Huawei. It is trained on the Ascend Atlas 800T A2 device and the Ascend MindSpore framework, and is the first state-of-the-art model trained entirely on domestically produced chips.
VidBee is an open-source video downloader built on the powerful yt-dlp engine. It helps users easily download audio and video content from over 1000 websites worldwide, including popular sites like YouTube and Bilibili, as well as lesser-known ones...
BabyVision is a multimodal understanding evaluation dataset launched by the UniPat AI team, evaluating the performance of multimodal language models (MLLMs) and image generation models on visual reasoning tasks. It includes two main tracks: MLLM evaluation and generation...
PixVerse R1, launched by PixVerse, is the world's first universal real-time world model, marking a significant breakthrough in AI video generation from "pre-recorded" to "real-time dynamic generation." R1 achieves this through three core technologies...
Engram, a conditional memory module developed by the DeepSeek team in collaboration with Peking University, enhances the performance of Large Language Models (LLMs). Engram stores static knowledge in dedicated memory by introducing a modern gram embedding table...
Baichuan-M3 is a new generation of open-source medical augmented language model officially released by Baichuan Intelligence. The model ranked first in diagnostic capabilities, medical illusion control, and Healthbench and Healthbench Hard benchmarks, surpassing...
Nuwax Agent OS is the world's first product-grade, open-source, general-purpose intelligent agent operating system. Nuwax Agent OS supports transforming AI from a simple conversational tool into an intelligent executor for enterprises, enabling autonomous execution...
Oh My OpenCode is an open-source enhancement plugin designed for AI programming assistants, enabling AI (such as OpenCode) to collaborate as efficiently as a real development team, thus improving the programming experience. The tool integrates various AI models (such as...)
Spirit-v1.5 is an embodied intelligence foundational model launched by Qianxun Intelligence. The model employs diverse and open data collection methods, abandoning the limitations of traditional "clean data," allowing the model to be exposed to richer actions and... during the pre-training phase.
Obsidian-skills is a set of open-source tools developed by the Obsidian team to help users better integrate AI tools (such as Claude Code) with the Obsidian note-taking system. It includes three core skills: obsidian...