Octo is an open-source AI-native team collaboration platform from Minglue Technology. It aggregates scattered AI agents into a unified space and enables efficient orchestration of people and agents, and agents and agents, through an IM format.
SemanticAudio is an audio generation and editing framework jointly developed by the Chinese University of Hong Kong, LIGHTSPEED, and Shanghai Jiao Tong University. The framework breaks down text-to-audio generation into two stages: 'semantic planning' and 'acoustic synthesis', at a higher level...
GeneBench-Pro is a research-grade benchmark from OpenAI, used to evaluate the ability of AI models to handle judgment-intensive analyses in computational biology. GeneBench-Pro includes 129 cross-genomics, quantitative biology...
WorldCupVoice is an open-source AI-powered real-time sports commentary system. By connecting to Agora RTC live streams, it uses a visual model to analyze the game footage and generate real-time audio commentary, which is then transmitted back to the live stream room for viewers to listen to simultaneously.
NVIDIA Nemotron-Labs-TwoTower is an open-source dual-tower architecture diffusion language model from NVIDIA, with approximately 60 bytes of total parameters and 3 bytes of active parameters. The model decouples context understanding and denoising generation into two independent towers, freezing the AR context...
LiveWorld is a generative video world model jointly developed by the University of Adelaide, the Australian National University, and other institutions. Its core function is to solve the problem of out-of-view dynamics; when an object leaves the camera's field of view, traditional models freeze its state...
Mirawork is a security-first desktop AI office assistant that supports macOS, Windows, and Linux. Users issue tasks using natural language, and Mirawork can automatically break down the requirements, invoke its built-in skill chain to complete the execution, covering documents...
Nano Banana 2 Lite is a lightweight AI image generation model developed by Google. It is a speed-up version that prioritizes localization and can generate a single image in 4 seconds, with a price of only $0.034 per thousand images.
LocateAnything is a visual language localization model from NVIDIA. Based on parallel bounding box decoding technology, it allows users to accurately select targets in an image by inputting natural language. The model supports multi-target detection, GUI localization, OCR text detection, etc.
Claude Sonnet 5 is the most capable agent model in the Sonnet series launched by Anthropic, supporting the creation of plans, the invocation of tools such as browsers and terminals, and autonomous operation.
yuxinlu1 Gemma4-12B is an open-source programming and agentic model series developed by individual developer Lu Yu based on the Google Gemma 4 12B instruction model. It includes the V1 code version and the V2 agentic version.
RedKnot is Xiaohongshu's open-source long-text LLM inference acceleration engine. It decouples the KV cache according to the attention header dimension, classifying it offline into global headers that need to be recalculated globally and local headers that can be reused locally. Combined with SegPagedAtten...
Huya VAM 1.0 (Vivid Avatar Model) is a real-time multimodal digital human basic model launched by Huya based on the DiT architecture. It can generate an AI digital human that can talk, sing and dance from a single photo.
Wan-Streamer is an open-source, end-to-end, real-time, full-duplex, multimodal foundational model from Alibaba DAMO Academy. It integrates the input and output tokens of text, audio, and video into a single causal sequence through a unified causal Transformer architecture...
LongCat-2.0 is a large-scale MoE language model open-sourced by Meituan, with a total of 1.6 trillion parameters. Each token activates approximately 48 billion parameters, and training and deployment are completed based on an AI ASIC supercomputing cluster.
VibeThinker-3B is a 3 billion parameter intensive inference model open-sourced by the Sina Weibo team. Built on Qwen2.5-Coder-3B, it utilizes an upgraded Spectrum-to-Signal post-training process, enabling it to perform well in verifiable mathematical and programming tasks...
Agent-Reach is a free and open-source AI agent internet capability scaffolding. The tool can install internet connectivity for mainstream agents such as Claude Code, Cursor, and OpenClaw with a single command, and supports Twitter...
BrowserBC is an open-source project released by Navers Lab under Einsia AI. Its core goal is to transform human actions in a browser into reusable natural language skills, enabling Web Agents to overcome...
FastContext is a lightweight code repository exploration model open-sourced by Microsoft, designed specifically for programming agents. The model decouples repository browsing from task solving; the main agent delegates tasks to FastContext via natural language queries...
Krea 2 is Krea AI's first base image generation model trained from scratch, positioned as an 'aesthetics-first' creative collaborator. It focuses on visual aesthetic consistency, style transfer, and creative control, rather than simply refining cue words...
HTML Anything is an open-source HTML editor for the Agent era from the nexu-io team. It includes 75 built-in Skill templates covering 9 delivery formats such as magazines, presentations, posters, and Xiaohongshu cards, and can automatically recognize local 18+ content...
DSpark is an open-source speculative decoding acceleration framework developed by DeepSeek in collaboration with Peking University. It specifically addresses the pain point of slow, incremental autoregressive generation of large models. It employs a semi-autoregressive generation architecture and utilizes a lightweight Markov header...
Penpot is an open-source design and code collaboration platform built on open standards such as SVG, CSS, and HTML, enabling UI design and interactive prototyping within a browser. The platform supports real-time multi-user collaboration, W3C Design Tok, and more.
GPT-5.6 is OpenAI's latest generation of large language models, currently only available in a 'limited preview' format to a few trusted partners due to US government regulatory requirements. This series is the first to adopt an astronomical naming system and offers three tiers...