News

Just moments ago, Alibaba's Qwen team open-sourced its latest generation of its flagship programming model, the Qwen3-Coder-480B-A35B-Instruct.The Qwen team says it's the team's most powerful open-source programming model for intelligentsia to date, with 480B parameter...

The kimi k2 technical report is finally here, perhaps the most recent technical report worth reading, full of sincerity, plus Manus two days ago the release of the build Agent lessons learned to eat better One sentence in a nutshell: Moonshot team with 1 trillion + parameter sparse...

As sales of Anthropic's artificial intelligence products surge, the company told investors that some of its profit-related metrics are improving, despite consuming a lot of cash. According to two people involved in the discussions, this financial performance has prompted some investors to say that if the company decides to move forward with financing...

Today at 1:00 am, Google released the first Gemini embedding model refreshed MTEB list record to become the first, and the price is very cheap per 1 million token as long as $0.15, has been open API. According to the results of Google's test on the multi-text embedding benchmarking platform MTEB...

According to foreign media TechCrunch combing reports, this year the United States a total of 36 technology startups successfully among the new unicorn camp, they are widely distributed in AI, biomedicine and other cutting-edge fields, each valuation exceeded the $1 billion mark. According to public information statistics, these 36 unicorns in this...

Halfway through 2025, are big models for text generation in the second half? Is OpenAI completely ignoring the API market? Is Grok3 not used at all? What is the future of the "big model wars"? Recently, Twitter blogger "karminski-dentist...

Every day, billions of users around the world use apps, websites, and digital interfaces built with Figma's support, such as Google Maps to navigate, Uber to take a cab, JetBlue to check in for a flight, Netflix to watch a show, Duolingo to learn a language,...

Microsoft CEO Satya Nadella shared on social media platforms, the latest release of Microsoft's groundbreaking healthcare AI system, MAI-DxO. MAI-DxO's biggest technological innovation is its model-agnostic design, which allows it to adapt to different vendors, different capabilities of language models, and...

Baidu Wenxin Big Model open source, as expected. Just today, Baidu officially announced that the Wenxin Big Model 4.5 series is officially open source, and also provides API services simultaneously. This time, Baidu launched a one-time 10 open source models, covering from 47B parameters of the Mixed Expert (MoE) model to lightweight 0.3B dense...

Overnight, Google Gemini CLI is completely on fire. It is an open source AI intelligence that can run on the terminal, and users can execute it directly on the terminal in the form of natural language commands. It also has the power of Google Gemini Pro 2.5...

On June 23rd, Galaxy General, a star company of humanoid robots, announced the completion of a RMB 1.1 billion financing, led by Ningde Times and Ningde Times' Puquan Capital, and followed by China Development Bank's CDB Sci-Tech, Beijing Robotics Industry Fund, and Jiyuan Capital. This is this year with the intelligent track...

China's AI industry in 2025 will seem to be back to the early stage of China's Internet in 2000, with new applications and products emerging all the time. But the difference is that we can no longer understand the new AI-driven world with the thinking of the Internet. In A...

On June 20, Shanghai-based medical AI startup Lianzhi Intelligence announced the completion of its Series A financing with a total scale of RMB 1 billion. The financing was co-led by Efonda Private Equity Fund Management Co. and Shangguo Investment Management, and co-funded by Shanghai Lianhe, Shengshi Capital, Baofeng Investment, Su Venture Capital, Junwei Venture Capital, Qingsong Capital, Zhangjiagang...

Early this morning, AI image generation top stream Midjourney released its first AI video generation model V1. V1 supports users to upload Midjourney-generated or external images for video generation, with two options for manual and automatic generation of action cues, which can be set to move the camera faster...

In the field of open source modeling, DeepSeek has brought another surprise. On the 28th of last month, DeepSeek came with a small update, upgrading its R1 inference model to the latest version (0528) and making the model and weights publicly available. This time, R1-0528 further...

In the first half of this year, the "big model six tigers" in the wisdom spectrum, MiniMax, zero one million things, Baichuan Intelligence, step leap star five companies, a total of 13 AI executive changes were exposed, of which 12 out, 1 new entry, and the first half of the domestic large manufacturers of AI talent changes in the number is not comparable to the number of large domestic manufacturers (domestic...

Early this morning, AMD hosted the 2025 Global AI Development Conference, which featured announcements and introductions of the latest AI chips and cloud infrastructure hardware devices. Sam Altman, co-founder and CEO of OpenAI, attended the conference as a special guest and, in conjunction with ADM, launched the I...

At 4am today, OpenAI announced the availability of the strongest model, o3-Pro, to all Pro users in ChatGPT and the API, which is already available as of today. o3-Pro automatically invokes a variety of tools, including the ability to search the web, analyze documents, reason about visual input, make...

Meta is in talks with Scale AI for a more than $10 billion investment. This would be the largest outside AI investment ever made by Meta, the parent company of Facebook, and one of the largest funding events ever for a private company. Publicly available information shows that Scal...

Early this morning, Alibaba open-sourced two new models of the Qwen3 series, Qwen3-Embedding and Qwen3-Reranker. These two models are designed for text representation, retrieval and sorting tasks, based on the Qwen3 base model training, fully inherited...