Kimi K2.6 is the latest flagship model from Dark Side of the Moon, boasting industry-leading code generation, long-term task execution, and agent clustering capabilities. The model has been tested in benchmarks such as Humanity's Last Exam, SWE-Bench Pro, and DeepSearch...
Qwen3.6-Max-Preview is an early preview version of the next-generation flagship large-scale model released by Alibaba's Tongyi Qianwen team. Compared to Qwen3.6-Plus, the model achieves significant improvements in three major dimensions: agent programming, world knowledge, and instruction compliance...
AI Fusion Video is an open-source, end-to-end AI video creation platform that uses an agent-based architecture to automate workflows from scriptwriting to video generation.
Fun-ASR 1.5 is a new generation version of the end-to-end speech recognition model launched by the Ali Tongyi team. A single model supports high-precision recognition of 30 languages, covering seven major Chinese dialect systems and more than twenty regional accents, with specific optimizations for classical Chinese poetry...
Lenovo's Tianxi AI Claw is an out-of-the-box AI assistant based on the OpenClaw architecture. It adopts a hybrid edge-cloud deployment model, supporting seamless collaboration across PCs, phones, and tablets. The tool requires no local environment configuration and can operate 24/7...
HyperFrames is an open-source AI-native video rendering framework from HeyGen. It adopts the 'Write HTML, Render video' principle and is designed specifically for AI agents and developers. The framework defines video as an HTML file, using data-* attributes...
Audio Flamingo Next is a next-generation audio language model (LALM) jointly open-sourced by NVIDIA and the University of Maryland. As the latest version in the Audio Flamingo series, it supports complex audio inputs up to 30 minutes in length and covers...
Voicebox is an open-source native speech synthesis tool, a cross-platform desktop application built on Tauri (Rust) and React. The tool offers voice cloning, text-to-speech, audio post-processing, and multi-track narrative editing features...
LingBot-Map is an open-source streaming 3D reconstruction model from Antminer Lingbo Technology. Ordinary RGB cameras can perform real-time camera pose estimation and scene 3D structure reconstruction during video acquisition.
Claude Opus 4.7 is Anthropic's latest flagship large-scale model, a direct upgrade from Claude Opus 4.6. The model excels in advanced software engineering tasks, achieving a SWE-bench Pro score of 64.3...
Qwen3.6-35B-A3B is an open-source hybrid expert (MoE) model launched by Alibaba's Tongyi Qianwen team. It has a total of 35 billion parameters and 3 billion activation parameters. The model focuses on extreme inference efficiency and agent programming capabilities, achieving high performance on multiple programming benchmarks...
GPT-Rosalind is a life sciences-specific inference model launched by OpenAI, named after Rosalind Franklin, the discoverer of the DNA double helix structure. The model is deeply optimized for 50 biological workflows and features hypothesis generation, experimental design, etc.
GPT-image-2 is OpenAI's next-generation native image generation model, rumored to be internally codenamed "Spud," and is currently undergoing grayscale testing on ChatGPT. The model is expected to be officially released in early April 2026 under codenames such as 'maskingtape-alpha'...
Relax is an open-source reinforcement learning training engine developed by the Xiaohongshu AI platform team, designed for large-scale model reinforcement learning in all-modal and agentic scenarios. It is built on Megatron-LM and SGLang, and adopts a service-oriented fault-tolerant architecture (Ray Service...).
Gemini 3.1 Flash TTS is Google's next-generation text-to-speech model, offering enhanced controllability, expressiveness, and sound quality. The model supports over 70 languages, incorporates audio tagging technology, and can precisely convert speech into speech using natural language commands...
HY-World 2.0 is an open-source multimodal world model launched by Tencent Hunyuan. It supports the generation of roamable 3D Gaussian sputtering (3DGS) scenes from text, single images, multiple views, or videos. The model uses four...
HappyOyster is an open-world model product launched by Alibaba's ATH Innovation Business Unit, positioned as an AI-generated experience platform capable of real-time construction and interaction. Based on world model technology, the product supports infinitely expandable real-time content generation...
StepAudio 2.5 TTS is a Contextual TTS (Context-Aware Speech Synthesis Model) launched by StepAudio, which for the first time introduces contextual understanding capabilities into the entire speech generation process.
Marble 1.1 is a next-generation generative world model from Fei-Fei Li's World Labs, primarily focused on image quality optimization. The model can transform single images, videos, or panoramic views into freely explorable, immersive 3D scenes, significantly improving lighting...
MAI-Image-2-Efficient is a self-developed texturing image model from Microsoft. It's a lightweight and efficient version of MAI-Image-2, designed for high-performance commercial mass production at a competitive price. While maintaining photorealistic image quality, it achieves a 41% cost reduction...
SearchClaw is a self-hosted AI deep research agent launched by RUC-NLPIR (Information Retrieval Laboratory, Renmin University of China), equipped with a web-based interactive interface. After a user submits a question, the system automatically performs multiple rounds of web searches...
Spark 2.0 is an open-source web-based 3D Gaussian sputtering rendering engine developed by World Labs (founded by Fei-Fei Li), built on Three.js and WebGL2. Spark 2.0 pioneered the LoD (Level of Detail) system, progressive streaming, and virtual memory...
ERNIE-Image is an open-source 8B parameter text-based image model from the Baidu Wenxin team. It is based on the Diffusion Transformer architecture and focuses on high controllability and accurate rendering of long texts.
OmniShow is a multimodal human-object interaction video generation model open-sourced by ByteDance in collaboration with the Chinese University of Hong Kong, Monash University, and the University of Hong Kong. As the first end-to-end framework supporting RAP2V (reference image + audio + pose), the model uniformly processes text,...