SlowFast-LLaVA-1.5 (SF-LLaVA-1.5 for short) is a high-efficiency video language model designed specifically for long video understanding. Based on a two-stream (SlowFast) mechanism, it balances processing more input frames with reducing the number of tokens per frame...
FutureX is a dynamic, real-time evaluation benchmark jointly released by research teams from ByteDance, Fudan University, Stanford University, and Princeton University, specifically designed for LLM agents' future prediction tasks. It utilizes a semi-automated pipeline to evaluate data from 195...
AskXiaoBaiO4 is China's first parallel thinking model, capable of simultaneously launching eight thinking paths, automatically selecting the optimal solution, and providing accurate answers. The model integrates Long-CoT reinforcement learning and process reward learning, possessing deep reasoning and high-quality thinking capabilities...
XBai o4 is an open-source large language model trained on "reflection-generated forms" and combining long CoT reinforcement learning and procedural reward learning. It performs well in complex reasoning and has surpassed OpenAI-o3-mini in medium mode.
DeepCode is a code generation platform based on multi-agent systems, developed by the Data Intelligence Laboratory at the University of Hong Kong. DeepCode can transform research papers, natural language descriptions, and other resources into high-quality, production-ready code, supporting various encoding methods...
VibeVoice is a new text-to-speech (TTS) model from Microsoft that generates expressive, lengthy, multi-speaker conversational audio, such as podcasts. The latest open-source VibeVoice-Realtime-0.5B model only requires 5...
EchoMimicV3 is a high-efficiency, multimodal, multi-task digital human video generation framework launched by Ant Group. The framework boasts 1.3 billion parameters, is based on a task-mixing and modality-mixing paradigm, and incorporates novel training and inference strategies to achieve fast...
SpatialGen is an open-source 3D scene generation model from GroupCore Technology. Based on a diffusion model architecture, it supports generating spatiotemporally consistent multi-view images from text descriptions, reference images, and 3D spatial layouts, and can further...
SpatialLM 1.5 is a powerful spatial language model developed by Qunhe Technology. Trained on a large language model, it can understand natural language commands and output spatial language containing spatial structure, object relationships, and physical parameters. Users...
WhisperLiveKit is an open-source real-time speech recognition tool that can transcribe speech into text in real time and supports speaker recognition. The tool is based on advanced technologies such as SimulStreaming and WhisperStreaming, providing ultra-low latency...
Wan2.2-S2V is an open-source multimodal video generation model that can generate cinematic digital human videos with a duration of up to minutes, requiring only a single still image and a piece of audio. It also supports various image types and aspect ratios.
Gemini 2.5 Flash Image (codename nano banana) is an AI image generation and editing model launched by Google. The model maintains consistency of characters across different scenes and supports precise image editing via natural language, such as...
Youtu-agent is an open-source intelligent agent framework developed by Tencent YouTu Lab, used for building, running, and evaluating autonomous intelligent agents. The framework achieves leading performance based on the open-source model DeepSeek-V3 and supports various model APIs and tools...
MiniCPM-V 4.5 is an edge-side multimodal model launched by Facewall Intelligence, featuring 8-bit parameters. The model excels in multiple fields including image processing, video processing, and OCR, achieving breakthroughs particularly in high refresh rate video understanding, capable of handling high refresh rate video...
Waver 1.0 is a next-generation video generation model launched by ByteDance. Based on the modified stream Transformer architecture, it supports text-to-video (T2V), image-to-video (I2V), and text-to-image (T2I) generation, and can be used in a single...
PixVerse V5, a self-developed AI video generation model from PixVerse Technology, has been launched globally. PixVerse V5 features comprehensive upgrades in dynamic effects, visual quality, consistency maintenance, and instruction compliance, enabling...
FramePackLoop is an infinite loop video generation tool based on FramePack. It combines a main video and linked videos into a looping video, suitable for use as a background, icon, or other video element.
The Wenxiaobai 5 is Wenxiaobai's flagship "All-in-One" large-scale model, boasting the highest level of intelligence among domestically produced large-scale models. The model has performed exceptionally well in multiple evaluations, such as a score of 64.7 on the AA-Index comprehensive evaluation index and high STEM ability scores...
Grok Code Fast 1 is an AI programming model launched by xAI, designed for fast and efficient basic coding tasks. The model can process 92 tags per second and has a 256k context window, making it suitable for rapid prototyping, code debugging, etc.
HunyuanVideo-Foley is an open-source end-to-end video audio generation model from Tencent's Hunyuan team. The model can generate high-quality audio effects that precisely match the video visuals based on the input video and text descriptions, solving the problems of existing AI video generation...
gpt-realtime is OpenAI's latest advanced speech model, designed specifically for real-world tasks. The model generates high-quality, natural speech, supports multiple languages and speech styles, and can understand non-verbal cues and adjust its speech according to the context...
OmniHuman, a 1.5-byte advanced AI model, can generate expressive digital human animations from single images and audio tracks. The model is based on dual-system cognitive theory, integrating a multimodal large language model and a diffusion transformer...
Meeseeks is a large model evaluation suite open-sourced by the Meituan M17 team, used to evaluate a model's ability to follow instructions. Meeseeks uses a three-level evaluation framework to comprehensively measure, from macro to micro, whether a model can strictly follow user instructions...
MAI-Voice-1 is the first highly expressive and natural speech generation model from Microsoft's AI team. The model can generate one minute of audio in less than one second on a single GPU, making it the most efficient speech system to date...