Search AI Products and News

Zhipu GLM-5.3 API launched, scoring 60 on the AA composite intelligence index, ranking among global frontier models alongside Claude Fable5 and GPT-5.6Sol, and tied with Kimi K3 as top open-source. Excels at complex coding, defensive cybersecurity, and long-horizon tasks.....

LG AI Research released K-EXAONE 2.0 on Hugging Face on July 31, the second generation of South Korea's sovereign AI project. It uses a hybrid attention MoE architecture with 750B total parameters and 37B active parameters, over 3x larger than the first version, making it South Korea's largest AI foundation model. Fully open-source under Apache 2.0, it's a rare move in Korean AI history.....

Anthropic CEO Dario Amodei denied ever proposing or advocating for banning open-source AI models from specific countries, amid rumors the US government may restrict such models. The statement sparked industry debate, with tech firms supporting open-source and some accusing Anthropic of seeking regulatory bans to curb rivals and protect its business interests.....

Silicon Valley's debate over open-source AI intensifies. OpenAI and Anthropic lobby for restricting Chinese open-source models citing national security, while Huang, Nadella, Musk, Zuckerberg, Pichai back open-source. China's rapid rise with models from Zhipu, Moonshot challenges US lead.....

After a hacker AI attack, Hugging Face tried using a US commercial large model API to analyze 17,000 logs but failed due to security mechanisms misjudging attackers vs. defenders. It then deployed the Chinese open-source model GLM5.2 on its own infrastructure, successfully completing massive log forensics. The incident highlights performance differences between AI models in security scenarios.....

On June 22, Hong Kong's AI sector surged. Zhipu, dubbed the 'global No.1 large model stock,' soared over 40% intraday to a high of HK$2,980, becoming the first HK stock firmly above HK$2,000. Its market cap topped HK$1 trillion, up over 2,000% since its January listing. The rally was fueled by large model rivalry and U.S. regulatory moves.....

The U.S. imposes export controls on Anthropic's AI models, restricting non-American users, intensifying tech competition and monopoly concerns. China's AI sector responds with openness; on June 13, Zhipu announced measures, highlighting diverging tech development paths.....

Google released the open-source experimental model DiffusionGemma on June 10, using a text diffusion architecture to achieve up to 4x faster text generation on dedicated GPUs compared to traditional autoregressive models, aiming to boost AI efficiency, though the company remains cautious.....

Google recently launched a Multi-Token Prediction (MTP) drafter for its open-source model Gemma4, leveraging speculative decoding architecture to boost inference speed by up to 3x while maintaining output quality and logical capabilities. Since its release, the model has seen rapid download growth, becoming one of the most popular open-source models globally.....

DeepSeek launches DeepSeek-V4 preview series as open-source, achieving million-character ultra-long context and leading domestic Agent collaboration and knowledge reasoning. Offers Pro (1.6T parameters, 49B activated) and Flash versions; Pro matches top closed-source models, Flash balances efficiency.....

GLM-5.1 introduces a breakthrough in long-range task handling, becoming the first open-source model capable of 8-hour continuous operation, shifting AI from simple Q&A to autonomous task execution.....

Google DeepMind releases Gemma4, a new open-source model with generational performance leap, now under Apache 2.0 license for commercial use. Four variants cater to mobile to workstation needs.....

Ant Group's Ring-2.5-1T, the world's first trillion-parameter reasoning model, excels in long-text generation, mathematical reasoning, and agent task execution. It reduces memory access by over 10x and boosts generation throughput by 3x for texts above 32K, significantly enhancing efficiency.....

GLM-5, a new open-source base model, achieves a paradigm shift from code generation to independently handling complex system engineering. It ranks fourth globally and first among open-source models on authoritative benchmarks, matching top-tier performance with exceptional engineering capabilities and a user experience close to Claude Opus4.....

Liquid AI launches LFM2.5, a compact foundational model series for edge devices and local deployment, featuring base and instruction-tuned versions with Japanese, vision, and audio variants. Optimized for CPU/NPU, it offers fast inference and is open-sourced on Hugging Face.....

Meituan's LongCat team open-sources LongCat-Video-Avatar, a video generation model that advances virtual human technology. It excels in long video generation, builds on LongCat-Video, and supports tasks like audio/text-to-video and video continuation, gaining developer attention.....

Meituan's LongCat team open-sourced the LongCat-Image model with 6B parameters, excelling in text-to-image generation and editing, aiming to bridge the gap between open and closed-source models by balancing performance and accessibility.....

MiroMind releases MiroThinker v1.0 with 256K context and 600 tool calls. It introduces 'Deep Interactive Scaling' for self-evolution via real-time feedback, integrating tools like search and code execution to autonomously handle complex tasks in hours.....

Kunlun AI Video Product SkyReels announced that the new version is expected to be officially launched in early November. Empowered by Kunlun's AI strategy, this product is expected to continue consolidating the company's leading position in the global AI video market and accelerate the realization of the industry vision that 'everyone can participate in professional video creation'. Currently, users around the world can log on to the SkyReels official website to apply for the test list and experience the model functions.

At 2025 Yunqi Conference, Alibaba Cloud CTO announced open-sourcing 300+ Tongyi Qianwen models with 600M+ downloads, showcasing AI leadership. Models span all sizes/modalities to boost industry applications.....