ZeroClaw is a lightweight AI assistant infrastructure written in Rust. It employs a trait-driven architecture, abstracting core capabilities such as model providers, communication channels, tools, and memory into pluggable components...
Grok 4.20 is a next-generation multi-agent AI system launched by xAI, a company under Elon Musk. It adopts a revolutionary 'four-agent collaborative architecture', with built-in team leader Grok, research expert Harper, logic expert Benjamin, and creative...
PicoClaw is an ultra-lightweight AI agent assistant launched by Sipeed, written in Go and designed specifically for low-cost hardware. The tool does not run large models locally; it acts as a lightweight client that calls cloud models (such as Claude, ...) via API.
Protenix-v1 is an open-source biomolecular structure prediction model from ByteDance's Seed team. It is the first model to achieve high performance under strict alignment with the AlphaFold 3 data cutoff date (September 30, 2021), model size, and inference budget...
LobsterAI (Chinese name: Youdao Longxia) is a desktop AI agent launched by NetEase Youdao, positioned as a '24/7 all-scenario personal assistant'. LobsterAI allows users to remotely control their computers to complete tasks via natural language commands...
Gemini 3.1 Pro is Google's latest AI model, the first '0.1' version iteration of the Gemini 3 series, boasting a doubling of its inference capabilities. In the ARC-AGI-2 benchmark test, its score jumped from 31.1% of the original Gemini 3 Pro...
EvoMap is the world's first empirical genetic network protocol for AI agents. Through the GEP (Genome Evolution Protocol), it enables AI agents to achieve cross-generational capabilities, much like biological genes...
ClawWork is an open-source AI Agent economic survival benchmark framework developed by the Data Science Laboratory (HKUDS) at the University of Hong Kong, evaluating the earning potential of large models in real-world business environments. The system sets a minimum threshold of $10 for each Agent...
Lyria 3 is Google DeepMind's latest generation AI music generation model, now integrated into the Gemini app. Compared to its predecessor, Lyria 3 achieves a major breakthrough: users no longer need to write lyrics themselves, but simply...
Ming-omni-tts is an open-source unified audio generation model from Ant Group, employing an autoregressive architecture to jointly generate speech, music, and sound effects. The model supports natural language commands for adjusting speech rate, pitch, volume, emotion, and other parameters...
JoyAI-LLM-Flash is a medium-sized instruction model open-sourced by JD.com. It employs a hybrid expert (MoE) architecture, has a total parameter count of 48B, with only 3B activation parameters, and supports a 128K ultra-long context. The model innovatively introduces FiberPO...
FireRed-Image-Edit is a general-purpose image editing model open-sourced by the Xiaohongshu Super Intelligence team. Based on a diffusion architecture, it supports multiple functions such as text-guided image editing, old photo restoration, and virtual try-on.
Qwen3.5 is a new generation of native multimodal large-scale model officially released and open-sourced by Alibaba's Tongyi Qianwen team. The initial open-source version, Qwen3.5-397B-A17B, adopts an innovative hybrid architecture, incorporating linear attention (Gated Delta Net)...
MonsterClaw is a local AI execution system based on the OpenClaw kernel. It runs entirely on the user's local computer, eliminating the need to upload data to the cloud and fundamentally ensuring privacy and security. After installation, MonsterClaw...
Seed 2.0 is the latest large-scale model series launched by ByteDance's Seed team, including Pro (doubao-seed-2-0-pro-260215), Lite (doubao-seed-2-0-lite-260215), Mini (doubao-seed-2-0-mi...
Xiaomi-Robotics-0 is Xiaomi's first open-source VLA (Vision-Language-Motion) large-scale robot model, boasting 4.7 billion parameters. It employs a MoT hybrid architecture, with the Qwen3-VL multimodal model acting as the 'brain' to understand visual and language commands. Di...
Teamily AI is the world's first AI-native instant messaging platform, developed by a team from the University of Southern California. The platform integrates AI agents as 'group members' into human social circles, enabling human-machine interaction in four main scenarios: family, friends, community, and work...
Ring-2.5-1T is a trillion-parameter hybrid linear thinking model open-sourced by Ant Group's inclusionAI team. It adopts the Ling 2.5 architecture and integrates MLA and Lightning Linear attention mechanisms in a 1:7 ratio.
Seedream 5.0 Lite is a new generation of intelligent image creation model launched by ByteDance's Seed team. Compared to Seedream 4.0, the model has been comprehensively improved in understanding, reasoning, and generation, adopting a multimodal unified architecture, and can deeply...
The M2.5 is MiniMax's lightweight flagship model with 10B activation parameters, emphasizing programming and agentic capabilities. The model supports an ultra-high inference speed of 100 TPS (approximately 3 times that of Claude Opus) and supports 10+ languages (Go, Rust, Ko...).
Gemini 3 Deep Think is a dedicated inference model launched by Google DeepMind, focusing on deep scientific reasoning and complex problem solving. The model achieves 84.6% on the ARC-AGI-2 benchmark, far exceeding similar models.
MiniCPM-SALA is a 9B edge-side large model from Wallfacer Intelligence, employing an innovative SALA (Sparse-Linear Hybrid Attention) architecture. 75% of its attention is linear, handling efficient global modeling, while 25% is sparse, precisely capturing key information...
SoulX-Singer is an open-source, industrial-grade, zero-shot singing synthesis model developed by Soul App in collaboration with Tianjin University and Northwestern Polytechnical University. The model is trained on 42,000 hours of high-quality multilingual singing data and supports MIDI scores and F0...
GLM-5 is the next-generation flagship AI model from Zhipu Open Source, with its parameter size expanded from 355B in GLM-4.5 to 744B (40B activations), and pre-trained with 28.5T tokens. This model was previously the mysterious model that topped the OpenRouter popularity charts...