Strix is an open-source, AI-driven security testing tool that helps developers and security teams quickly discover and verify vulnerabilities in applications. The tool reduces false positives by dynamically running code and simulating real hacker attacks.
IndexTTS2, a new text-to-speech (TTS) model developed by Bilibili's voice team, has been officially open-sourced. The model has achieved significant breakthroughs in emotional expression and duration control, and is the first autoregressive TTS model to support precise duration control...
AgentCLUE-ICabin is an AI agent benchmark focused on automotive smart cockpit scenarios, comprehensively evaluating the tool invocation capabilities of large language models within the smart cockpit. The benchmark is built upon 12 common vehicle usage scenarios, covering everything from daily...
MobileLLM-R1 is a series of high-performance inference models from Meta, designed specifically for mathematical, programming, and scientific reasoning. The series includes base models and final models, with versions containing 140 million, 360 million, and 950 million parameters respectively. The models are not universally applicable...
veCLI is a command-line AI tool launched by ByteDance's Volcano Engine, improving development efficiency. It seamlessly integrates with multiple models, including the Doubao large model 1.6, allowing developers to quickly generate code and complete local deployment through natural language interaction...
ZipVoice is a high-efficiency zero-shot speech synthesis (TTS) model released by Xiaomi AI Lab. The model is based on the Flow Matching architecture and includes ZipVoice (single speaker) and ZipVoice-Dialog (dialogue)...
PP-OCRv5 is a high-efficiency and accurate text recognition model launched by Baidu. The model is based on a two-stage processing flow and is specifically designed for fast and accurate detection and recognition of text in images.
Stable Audio 2.5 is the latest audio generation model from Stability AI, designed specifically for enterprise-level sound production. The model features rapid generation (three minutes of audio in just two seconds), dynamic music creation, and audio restoration capabilities.
The Live Interpreter API is a new feature in Azure Speech Translate that enables real-time multilingual speech translation. Without requiring users to manually set the input language, the Live Interpreter API automatically and continuously recognizes the language being used...
Youtu-GraphRAG is an open-source graph retrieval enhancement and generation framework from Tencent YouTu Lab. By organizing knowledge into a graph and combining it with a large language model for retrieval and reasoning, it helps models answer complex questions more accurately and reduces irrelevant information.
ERNIE-4.5-21B-A3B-Thinking is a large-scale language model launched by Baidu, focusing on reasoning tasks. It adopts a hybrid expert (MoE) architecture, with a total of 21 billion parameters, each token activating 3 billion parameters, and supports long strings of up to 128K...
FunAudio-ASR is an end-to-end speech recognition model developed by Alibaba DAMO Academy, designed to solve key problems in enterprise deployment. Through its innovative Context enhancement module, it effectively optimizes against "illusion" and "language cross-referencing"...
GPT-5-Codex is a model from OpenAI specifically optimized for programming, further enhanced from GPT-5. The model focuses on real-world software engineering tasks, such as building projects from scratch, code refactoring, debugging, testing, and code...
ROMA (Recursive Open Meta-Agent) is an open-source multi-agent system framework from the Sentient AGI team. It uses a recursive hierarchical structure to break down complex tasks into parallelizable subtasks, which are then assigned by parent nodes to child nodes for execution...
Grok 4 Fast is a fast version of the AI model launched by xAI. Its biggest feature is its fast response speed, up to 10 times faster than the standard version, generating 75 tokens per second. It can quickly complete simple queries, basic code generation, and other tasks...
Mini-o3 is an open-source model jointly developed by ByteDance and the University of Hong Kong, designed specifically to solve complex visual search problems. Through reinforcement learning and image-based tools, the model can perform deep, multi-turn inference, with the number of inference rounds adjustable...
Lumina-DiMOO is a next-generation multimodal generation and understanding model open-sourced by the Shanghai Artificial Intelligence Laboratory and other institutions. The model employs a fully discrete diffusion architecture to uniformly process multimodal data such as text and images, supporting text-to-image generation...
UnifoLM-WMA-0 is an open-source world model and motion architecture for multiple robot types, designed specifically for general robot learning. Its core is the world model, which understands the physical interactions between the robot and its environment and possesses simulation capabilities...
InfiniteTalk is a new digital human-driven technology launched by Meituan's Visual Intelligence Department. Through a sparse frame video dubbing paradigm, it requires only a small number of keyframes to drive the digital human to generate natural and smooth videos, solving the problem of speech dubbing in traditional technologies...
ReSum is a new WebAgent inference paradigm launched by Alibaba Tongyi, belonging to the Tongyi DeepResearch family. It addresses the context length limitation problem faced by WebAgents in long-view-of-fact tasks by periodically refining the interaction history...
LLaSO (Large Language and Speech Model) is the world's first fully open-source speech model, developed by Beijing DeepLogic Intelligence Technology Co., Ltd. It addresses long-standing issues in the field of Large Language and Speech Models (LSLMs)...
Nano Bananary is an open-source image editing tool developed based on the Google Gemini image model. It supports a Chinese interface and brightness/darkness theme switching, offering over 50 image transformation effects without complex prompts...
PromptEnhancer is an open-source suggestion rewriting framework developed by Tencent's Hunyuan team for improving text-to-image (T2I) models. It utilizes Chain-of-Thought (CoT) suggestion rewriting and a dedicated reward model, AlignEval...
VLAC is an embodied reward model released by the Shanghai Artificial Intelligence Laboratory. Based on the InternVL multimodal model, it integrates internet video data and robot operation data to provide reinforcement learning for robots in the real world...