Composer 1.5 is Cursor's self-developed AI programming model, trained by scaling up reinforcement learning by 20 times. It's a 'thinking' model that dynamically generates a thinking token inference code library, supporting rapid responses to simple questions...
iFlytek Spark X2 is a new generation of large-scale models trained by iFlytek based on domestically produced computing power. It adopts a 293B MoE sparse architecture, and its general capabilities are comparable to the top international level. Its mathematical reasoning, 130+ multi-language capabilities and other capabilities are comparable to the best international standards.
OpenStoryline is Xiaohongshu's first open-source video editing agent with directorial thinking, enabling intelligent creation throughout the entire process from material retrieval and script writing to rendering output through natural language dialogue.
Clawra is an open-source AI virtual companion based on the OpenClaw framework. Clawra has a complete persona: a former K-pop trainee, currently an intern in the San Francisco market.
RynnBrain is an open-source embodied intelligent brain model launched by Alibaba DAMO Academy, enabling robots to possess spatiotemporal memory and physical spatial reasoning capabilities for the first time. The model is trained based on Qwen3-VL and uses the self-developed RynnScale architecture...
Seedream 5.0 is ByteDance's next-generation AI image generation model. For the first time, the model supports online image retrieval, more accurately understands abstract prompts, and generates 2K/4K high-resolution images. The model also adds brush editing functionality, supporting...
Qwen-Image-2.0 is a new generation image generation model launched by Alibaba's Tongyi Qianwen platform, supporting two core capabilities: accurate text rendering and realistic, detailed texture. The model supports direct output of professional infographics, PPT presentations, and more using 1k token long commands.
HY-1.8B-2Bit is Tencent's first industry-grade 2-bit edge-side large-scale model launched by Hunyuan. The model is based on the 1.8B model and compressed using Quantization-Aware Training (QAT). HY-1.8B-2Bit has an equivalent parameter count of only 0.3B and a size of only 300MB...
Riverflow 2.0 is a production-grade image generation and editing model from Sourceful, designed specifically for marketing and creative teams. The model includes two versions: PRO and FAST. PRO prioritizes ultimate quality and consistency, excelling in text rendering, ...
PaperBanana is an automated academic illustration generation framework jointly developed by Peking University and Google Cloud AI Research, addressing the pain point of time-consuming and labor-intensive data creation for AI researchers in academic papers. The system employs an innovative multi-agent approach...
Nanobot is an ultra-lightweight personal AI assistant open-sourced by the Data Intelligence Laboratory at the University of Hong Kong. With only about 4,000 lines of code, it fully replicates the core functionality of the OpenClaw agent. Nanobot features web search, file operations, and more...
Seedance 2.0 is a next-generation AI video generation model launched by ByteDance's JiDream, emphasizing multimodal reference and efficient creation capabilities. The model supports comprehensive reference of the first and last frames, video clips, and audio, accurately replicating camera movement logic...
GPT-5.3-Codex is OpenA's flagship intelligent coding model, which combines programming capabilities with professional reasoning knowledge, increasing speed by 25% and halving token consumption.
Claude Opus 4.6 is Anthropic's flagship AI model, an upgrade from Claude Opus 4.5. The model is the first to support an ultra-long context window of 1 million tokens, leading the way in programming, inference, and complex task processing.
LingBot-VA is the world's first causal video-action world model for general robot control, open-sourced by AntLingbo Technology. The model unifies video world modeling and policy learning into an autoregressive framework, enabling the robot to simultaneously predict...
Kilo CLI 1.0 is an open-source command-line tool from Kilo Code, designed specifically for agent engineering. Built on OpenCode, it supports over 500 AI models and allows developers to...
CL-bench is a context learning capability evaluation benchmark jointly launched by Tencent Hunyuan and Fudan University. It can measure the ability of a large language model to learn and apply knowledge from new contextual information in real time.
MiniCPM-o 4.5 is Wallfacer's open-source 9B parameter full-modal flagship model, which adopts an end-to-end architecture that integrates SigLip2, Whisper, CosyVoice2 and Qwen3-8B.
Keling AI 3.0 is Kuaishou's new generation multimodal AI creation model, enabling an 'All-in-One' native creation workflow. Model version updates include the introduction of Video 3.0, supporting AI intelligent scene creation, 15-second long video generation, multilingual audio...
ACE-Step 1.5 is an open-source music generation model jointly developed by ACE Studio and StepFun, enabling commercial-grade music generation on consumer-grade hardware. The model employs a hybrid architecture, with a language model acting as a planner to incorporate user prompts...
Xuanwu CLI (xw-cli) is a domestically developed large-scale model deployment tool open-sourced by Qingmao Intelligence. It is deeply optimized for domestic chips such as Huawei Ascend, Muxi, and Suiyuan, and through automatic hardware detection and intelligent engine scheduling, it allows users to deploy large models without complex configurations...
SoulX-FlashTalk is the first 14B-parameter real-time digital human generation model open-sourced by the AI team under Soul App, achieving a sub-second latency of 0.87 seconds and a high frame rate of 32fps.
Qwen3-Coder-Next is an open-source programming intelligent agent model from Alibaba's Qwen team. It adopts a hybrid expert (MoE) architecture with a total of 80B parameters, but only 3B are activated in each inference, which greatly reduces the cost of GPU memory and computing power.
GLM-OCR is a lightweight multimodal OCR model open-sourced by Zhipu AI. With only 0.9B parameters, it topped the OmniDocBench V1.5 leaderboard with a score of 94.6.