SALMONN-2 is a general-purpose audio language model jointly open-sourced by Tsinghua University, Shanghai Artificial Intelligence Laboratory, and the University of Cambridge. It is built upon ByteDance's SPEAR unified self-supervised audio encoder and multi-layer feature fusion adapter, and...
Wan3.0 is the latest video generation model launched by the Alibaba Cloud Wanxiang team. The model has been comprehensively upgraded in dimensions such as generation duration, versatile creation, all-around reference, and realism. It can generate 30-second videos in a single run and supports doc, xls, and other formats for the first time.
Hunyuan3D-Buffalo 1.0 is a unified 3D multimodal framework launched by Tencent Hunyuan. It integrates 3D question answering, spatial positioning, text-based 3D, instruction editing, and component generation into a single process through the shared Hunyuan3D-VLM backbone.
InstructAV2AV is an audio and video co-editing model jointly developed by the Beijing Academy of Artificial Intelligence and Peking University. Users only need a single natural language command to simultaneously edit video footage and corresponding content during the end-to-end generation process...
JoyAI-Video-Edit is a real-time streaming video editing model developed and open-sourced by JD.com. Based on a 16B-parameter autoregressive diffusion architecture, it achieves an inference speed of 30FPS at 720P resolution and supports stable streaming editing of any duration.
SeedRealtime is a native full-duplex audio and video model launched by ByteDance's Seed team. Based on a unified architecture, it natively integrates audio, video, and text to achieve real-time multimodal interaction, allowing users to watch, listen, and speak simultaneously.
MAGI-2-preview is Sand.ai's open-source MoE multimodal video generation model, boasting a total of 114 bytes of parameters and only 6 bytes of activation. The model continues the single-stream architecture, integrating text, video, and audio into a single T...
Shieldstral is an open-source, multimodal content security classification model with 3B parameters, developed by Mistral AI and built upon Ministral-3B. The model redefines traditional fixed-category content moderation as a binary question-answering task...
MemHarness is an LLM agent memory reconstruction framework developed by the Shanghai AI Lab in collaboration with universities such as Zhejiang University, Fudan University, and Shanghai Jiao Tong University. Existing memory enhancement agents often directly inject retrieved historical experiences into the context, but if old experiences...
Hy ASR 3.0 preview is a next-generation speech recognition model launched by Tencent Hunyuan. Based on the Hy3 large language model, it integrates high-precision speech recognition with deep semantic understanding. The model supports Mandarin, English, Cantonese, and 10 major dialect regions...
Orchard is an open-source agentic AI modeling framework developed by Microsoft Research. Its core is the Orchard Environment service based on Kubernetes, supporting cross-domain sandbox reuse for data distillation, reinforcement learning rollout, etc.
ForgeStencil is the world's first AI optimization system supporting automated Stencil research and deployment, jointly developed by Wallfacer Intelligence and the OpenBMB open-source community. It is based on a dual-agent closed-loop architecture consisting of a Kernel Agent and an App Agent...
E-Bench is a multi-step tool usage evaluation benchmark launched by Tencent's Hunyuan team in collaboration with Tsinghua University AIR and Southeast University. It is based on three real-world products—Honor of Kings, QQ Music, and Tencent Meeting—and constructs a fully synthetic virtual environment...
SenseNova U1.5-Lite-Preview is a lightweight, native unified multimodal model preview version open-sourced by SenseTime. It is based on the NEO-Unify architecture iteration and integrates visual understanding, reasoning, generation, and editing with only 8B-MoT parameters.
MindMemOS is an open-source AI Agent memory operating system from Huawei Noah's Ark Labs, designed to address the pain point of agents forgetting their memories after use. The system decouples memory from individual agents into independent assets that can be reused across frameworks...
Qwen 3.8-Max is the flagship large-scale model launched by the Alibaba Cloud Qwen team, with a total of 2.4 trillion parameters. It is an extension of the Qwen 3.5 architecture. As the most powerful model in the Qwen family to date, it is the first to be open-sourced...
FeyNoBg is an open-source automatic background removal model from Feyn Labs. It's based on the BiRefNet architecture with significant improvements and boasts 263 million parameters. The model achieved excellent results in 4 out of 8 benchmark tests, including UHRSD-TE and HRSOD-TE...
Qwen-Audio-3.0-ASR-Flash is a large-scale speech recognition model launched by Alibaba's Qianwen team. It is now available for use through Alibaba Cloud's Bailian platform, providing three versions: Flash, Filetrans, and Streaming.
MiniMax H3 is a general-purpose, multi-modal generative model developed by Xiyu Technology. It breaks down traditional task and modal boundaries, supporting unified understanding and native generation of text, images, video, and audio. The model can output native dual-channel audio and video...
Codex Security CLI is an open-source AI code security scanning CLI tool from OpenAI, formerly a closed-source security plugin built into Codex. The tool automatically scans for code vulnerabilities based on large-model semantic understanding and deeply verifies the authenticity of attack paths...
AngelSpec is an open-source end-to-end speculative decoding training framework developed by Tencent's Hunyuan team. Built on TorchSpec, it supports six draft architectures, including MTP autoregressive and DFly block parallelism. Inference and training are decoupled; the inference engine...
Grok Voice Think Fast 2.0 is an end-to-end speech-to-speech model from SpaceXAI (xAI). It employs a native speech-to-speech architecture, directly understanding and responding without requiring a multi-stage process of recognition, reasoning, and synthesis.
Qwen-Audio-Agent is an open-source real-time speech agent framework launched by Alibaba's speech AI team, built on the Qwen-Audio-3.0-Realtime model. The framework serves as a unified real-time speech entry layer, allowing users to interact with the system through natural language processing...
Metasequoia Input Method is a free and open-source Chinese input method based on the native Windows TSF framework, designed specifically for Win10/Win11. Unlike most RIME-based solutions, this tool is entirely self-developed and natively compatible with UWP and high-DPI...