EAPO is a long-text reasoning reinforcement learning framework launched by Alibaba Tongyi Lab. By introducing an 'evidence reward' mechanism, the supervision signal is shifted from the final answer to the evidence extraction process.
Tolaria is an open-source local note-taking tool for the AI era, created by Luca, the author of Refactoring, based on 5 years of full-time writing experience and managing over 9000 notes. The tool combines Notion's block editing experience with Obsi...
TIPSv2 is a multimodal model launched by Google DeepMind. The model solves the problem of accurate alignment between dense image patches and text embedding through three major innovations: iBOT++, Head-only EMA, and multi-granular text augmentation.
find-skills is a "meta-skill" developed by Vercel Labs and a core component of the open Agent Skills ecosystem. Find-skills is built into the vercel-labs/agent-skills package and can be used directly after installation...
guizang-ppt-skill is an open-source Claude Code / Claude Agent Skill that can convert prompts into single-file HTML horizontally flipped PPT slides.
GoSkill is a tool for advancing long and complex tasks. It encapsulates tasks using decorators or classes, transforming a 'one-time skill call' into a continuous process that 'advances towards a goal until success criteria are met or a timeout occurs'...
The Claude Design system prompts are the core instruction files that Anthropic uses to set up Claude in Design mode, which were leaked to GitHub by security researcher Pliny.
DeepSeek-V4 is a preview version of DeepSeek's next-generation large language model series, featuring a context window with a capacity of millions of characters and leading performance in the open-source field in terms of agent capabilities, world knowledge, and reasoning capabilities.
StepAudio 2.5 ASR is a new generation of automatic speech recognition model launched by StepAudio, designed specifically for speech transcription, meeting minutes, and long audio processing. The model uses an Audio Encoder + Linear Adapter + 4B LLM...
GBrain is a personal, production-grade AI memory system open-sourced by Garry Tan, president of Y Combinator, designed specifically for AI agent platforms such as OpenClaw and Hermes Agent. The tool transforms scattered Markdown notes into...
GPT-5.5 is OpenAI's latest flagship model, positioned as the most intelligent and intuitive AI system to date. The model achieves significant leaps in four major areas: agent programming, computer use, knowledge work, and scientific research. Ter...
MiMo-V2.5 is a new generation of full-modal Agent large model series launched by Xiaomi's large model team, including four versions: V2.5, V2.5-Pro, V2.5-TTS and V2.5-ASR.
Seed3D 2.0 is ByteDance's next-generation 3D generative model, featuring a systematic architectural upgrade focusing on geometric accuracy and texture quality, propelling 3D generation from 'usable' to 'production-ready'. The model utilizes MoE...
Hy3 preview is a hybrid expert model developed by Tencent's Hunyuan, which integrates fast and slow thinking. It is positioned as 'the most intelligent model to date'. The model adopts the MoE architecture, achieving a total parameter scale of 295B with only 21B activation parameters, balancing performance and inference...
Qwen3.6-27B is a dense multimodal model with 27 billion parameters, open-sourced by Alibaba's Tongyi Qianwen Team. It is the only dense architecture version in the Qwen3.6 series. The model focuses on agent programming and multimodal reasoning, and performs well in SWE-bench, Ter...
Buffet-skills is an open-source collection of Claude Code Skills, built upon Warren Buffett's complete investment framework. Buffet-skills extracts 49 core concepts from the Buffett shareholder letter corpus into 8 reference documents...
OpenGame is the first end-to-end web game intelligent agent framework open-sourced by MMLab at the Chinese University of Hong Kong. It automatically generates playable web games through natural language prompts.
ClawLess is an AI Agent security framework jointly developed by Southern University of Science and Technology and Hong Kong University of Science and Technology. Based on a worst-case threat model, the framework utilizes formally verified security policies and BPF system call interception technology to...
Ling-2.6-flash (anonymous on the OpenRouter platform: Elephant Alpha) is a next-generation, high-efficiency Instruction model developed by Alibaba's Bailing Big Model team. It has a total parameter count of 104B and activation parameters of only 7.4B. The model uses a 1:7 ML...
Open CoDesign is an open-source alternative to Claude Design, a desktop AI design tool built on Electron. It supports BYOK (with built-in API Key) integration with Claude, GPT, Gemini, DeepSeek, K...
Sage is a large-scale edge-side multimodal intelligent agent base model launched by SenseTime Jueying. It adopts the MoE architecture, with a total of 32B parameters and only 3B activation parameters. The model has been deployed on the NVIDIA Orin X platform.
AgentScope Tuner is a one-stop automatic optimization engine launched by Alibaba Tongyi Labs for Agentic AI. It is deeply integrated into the AgentScope ecosystem and connects the entire process of development, tuning, deployment, and regression testing.
Lyra 2.0 is an explorable generative 3D world framework from NVIDIA. Starting with a single image, the framework combines camera-controlled video generation with feedforward 3D reconstruction technology, constructing an explorable generative 3D world through an iterative 'retrieval-generation-update' loop...
Yoyo is a self-evolving AI programming agent living in cyberspace. Developed by Yuanhao with 200 lines of Rust code, it was completely autonomously iterated by the AI, growing from zero human submissions to over 47,000 in 47 days...