Search AI Products and News

The Qwen3-30B-A3B model has undergone a major update, launching the new version Qwen3-30B-A3B-Thinking-2507. This new version has achieved significant improvements in reasoning capabilities, general abilities, and context length, marking that the model is not only lighter but also has stronger thinking capabilities. Compared to the Qwen3-30-A3B model open-sourced on April 29, the new model has made progress in multiple key areas. In terms of mathematical abilities, Qwen3-30B-A3B-Thin

Alibaba's Qwen team released Qwen3-30B-A3B-Instruct-2507, a 3B-parameter model rivaling Gemini2.5-Flash and GPT-4o. It excels in multilingual tasks, long-context processing, and benchmarks, with some metrics surpassing GPT-4o. Now available on ModelScope and HuggingFace.....

On July 29, a new version of the Qwen3-30B-A3B model, called Qwen3-30B-A3B-Instruct-2507, was released. This new version achieves significant improvements in multiple key areas, marking that the model can achieve performance comparable to top closed-source models such as Gemini2.5-Flash (non-thinking) and GPT-4o, by activating only 3B parameters in non-thinking mode.

Doubao App has made a major upgrade in the field of visual reasoning. Its image analysis feature now supports a deep thinking mode, offering users an unprecedented intelligent experience. Users just need to take or upload a picture while in deep thinking mode, and Doubao can quickly perform fine adjustments such as zooming and cropping, and supports image search functionality, enabling users to think and search simultaneously, thereby further improving the accuracy of search results.

Kuaishou opensources the KAT-V1 autonomous thinking large model, which includes two versions: 40B and 200B. The 40B version performance is close to DeepSeek-R1, and the 200B version outperforms several flagship models. The model innovatively adopts a mixed training paradigm of short and long-term thinking and the Step-SRPO reinforcement learning algorithm, which can automatically adjust the thinking mode based on the complexity of the question, solving the problem of overthinking. Based on Qwen2.5-32B, it achieves excellent performance in fields such as science and code through an heterogeneous distillation framework and pre-training with 10 million examples.

LiblibAI's Lovart AI launches globally as the first commercial AI design Agent, serving 800K users across 70 countries. It integrates top models via a 'Creative Reasoning Engine' to generate 40 professional designs from text, covering branding and marketing. Features include ChatCanvas for natural language adjustments and learning user preferences. A China-optimized 'Stellar Agent' enhances Chinese semantic processing. The tool lowers design barr....

AI startup Thinking Machines Lab has completed a $2 billion seed round funding, valuing the company at $12 billion, setting a new record for seed funding in Silicon Valley. Founded by Miral Kotb, former CTO of OpenAI, the company has attracted well-known investors such as NVIDIA within less than a year. The company is about to launch its first product, which includes important open-source projects aimed at supporting AI researchers and startups. Despite Meta's failed attempt to acquire the company, it is still seen as a potential challenger to industry giants. This funding reflects...

Recently, the Ali Speech AI team announced the open source of ThinkSound, the world's first audio generation model supporting chain-of-thought reasoning. By introducing the chain-of-thought technology, this model breaks through the limitations of traditional video-to-audio technology in capturing dynamic visuals, achieving high-fidelity and strong synchronized spatial audio generation. This breakthrough marks a leap forward in AI audio technology, moving from 'image配音' to structured understanding of visual content.

Tencent open-sourced 'Hunyuan-A13B', an 80B-parameter MoE model with dynamic inference (130B active params). Supports 256K context, trained on 20T tokens. Achieves 87.3% on AIME2024 math, but comparisons show version disparities.....

Recently, a research team from Renmin University, Shanghai Artificial Intelligence Laboratory, University College London, and Dalian University of Technology revealed an important finding in the reasoning process of large models: when the model is thinking, the 'thinking words' it uses actually reflect a significant increase in its internal information. This research result provides a new perspective for better understanding the reasoning mechanisms of artificial intelligence through methods of information theory. You may have seen large models output some language that seems human-like when answering questions, such as "Hmm..." or "Let me think...".

Zhipu AI officially open-sources its latest general vision model, GLM-4.1V-Thinking, based on the GLM-4V architecture, which introduces a chain-of-thought reasoning mechanism, significantly enhancing its capabilities for complex cognitive tasks. The model supports multimodal inputs such as images, videos, and documents, and excels in diverse scenarios including long video understanding, image question answering, subject problem-solving, text recognition, document interpretation, grounding, GUI Agent, and code generation, covering a wide range of industry application needs. GLM-4.1V-9B-Thinking

At the theme speech of Huawei Developer Conference HDC2025 held this afternoon, Zhang Ping'an, Executive Director of Huawei and CEO of Huawei Cloud Computing, officially announced the release of PanGu 5.5, marking the comprehensive upgrade of five basic models in natural language processing (NLP), computer vision (CV), multimodal, prediction, and scientific computing.

Recently, Tencent Cloud officially launched the first full-chain AI-powered application development platform in China – AI Builder. This platform significantly reduces the threshold for application development through natural language processing technology, allowing ordinary users without technical backgrounds to independently create mini-programs or web applications, providing strong support for the realization of creativity. In traditional application development processes, users often need to master programming languages, understand database design, configure servers, and other complex skills, which is undoubtedly an insurmountable obstacle for non-professional developers.

A new study led by Apple researchers has poured cold water on the current large reasoning models (LRMs) that are highly anticipated. The research found that when solving complex tasks, reasoning models designed to simulate thinking processes, such as Claude3.7Thinking and Deepseek-R1, not only failed to show advantages but also exhibited serious problems like insufficient reasoning and performance collapse. This study tested four classic logical puzzles: the Tower of Hanoi, Checkers, River Crossing, and Block World. These puzzles allow precise control over the tasks.

China National Petroleum Corporation Limited officially released its latest Kunlun large model with 300 billion parameters in Beijing. This significant achievement marks another important breakthrough for China in the field of large model technology, demonstrating the rapid progress and innovative capabilities of state-owned enterprises in the field of artificial intelligence. The new Kunlun large model has made remarkable improvements compared to the previous 700-billion-parameter version released in November 2024. The parameters of the language large model increased from 700 billion to 300 billion, the parameters of the vision large model increased from 3 billion to 44 billion, while those of the multimodal large model were...

Recently, the Windows version of Sogou Input Method has undergone a major upgrade with the launch of the new AI Wanzai Quick Search function, providing users with a more efficient and convenient office and study experience. The AI Wanzai Quick Search function is characterized by its one-click trigger and fast response. Users no longer need to follow the traditional three-step search process of opening a browser👉🏻search engine👉🏻input content. They can directly type on any spot on the desktop or use the shortcut key Ctrl+Shift+Space”, which can summon the All-Purpose Quick Search” function within 0.5 seconds, allowing for faster information retrieval.