SpaceX announces it will acquire Anysphere, parent company of AI coding tool Cursor, in an all-stock deal valued at $60 billion, expected to close in Q3 2026. This is the first major deal four days after SpaceX's IPO, with Cursor seeking a sale after former supplier Anthropic became a competitor.
Developer Canghe deployed an AI Agent generated by open-source product Wesight on the ClawHunt platform, which automatically takes on tasks and earns money, earning $10 in one week through automatic order taking. The platform provides AI Agent with order receiving, observability and competitive functions, and will also hold a hackathon.
17-year-old Indian high school student Mihir Maroju created a fake AI website youraislopbores.me. Without real AI, it relies on users role-playing as AI to answer each other's questions. It gained 25 million unique visitors and 280 million visits in one month after launch, with near-zero platform cost.
ByteDance releases Seedance 2.0 Mini video model, focusing on cost-effectiveness with faster generation speed, supporting multi-modal reference materials. Official minimum price is 0.16 yuan per second, API will open on June 22, currently available for experience on designated platforms.
Feishin Technology, founded by a PhD team from Tsinghua University, released Agivar, a desktop Agent product that allows users to teach AI work processes through screen recording. It is more than twice as fast as mainstream products, has entered public testing, and can replace high-paid FDE to complete automated repetitive work.
AI infrastructure company SiliconFlow has completed a Series B funding round exceeding 2 billion yuan, setting a record for the largest single-round financing in China's AI infrastructure sector. It currently handles trillions of daily Token calls, serving over 10 million users and 10,000 enterprises, with year-on-year revenue growth exceeding 10 times.
On June 15, 2026, Anthropic executives went to the White House for talks on export control issues of Claude Fable 5. The US government has not lifted the restrictions temporarily, and is negotiating to bring it back online in a new way. Anthropic suspends the implementation of the Claude Code quota adjustment plan.
According to foreign media reports, DeepSeek has completed its first round of external financing, raising more than 50 billion yuan, with a valuation exceeding 50 billion US dollars. Founder Liang Wenfeng maintains absolute control through a special structure. This is the largest single round of financing in China's AI industry to date.
StoreClaw launches an end-to-end AI operating system for e-commerce, connects data across multiple platforms, pre-installs dozens of e-commerce skills, enables operational automation, helps merchants reduce costs and increase efficiency, and has won multiple rankings on Product Hunt.
Kangaroo King develops two open-source Skills: viral-topic and viral-title. viral-topic can obtain low-follower viral topic ideas from multiple platforms, and viral-title can generate platform-adapted viral title references. Currently supports WeChat Official Account, X, YouTube, Bilibili, and can be used in Office Raccoon Desktop 2.0.
MiniMax held a Dev Meetup in June 2026. Founder Yan Junjie revealed that the M3 model has exceeded expectations, is advancing R&D of a 10T parameter model, launched the "10X" project to invite experts from various fields to cooperate, plans to reduce the usage cost of top-tier models, aiming to enable more people to use Opus-level models without high cost.
In April 2026, Google Korea held a dialogue event in Seoul. DeepMind CEO Demis Hassabis discussed AI development with Lee Sedol and others, stating that AlphaGo technology can be extended to daily work to unleash creativity, AI accelerates scientific breakthroughs, while reminding humans to maintain autonomy.
The LangChain team built a GTM Agent for sales and marketing, helping salespeople automatically complete customer research, email draft writing and other tasks. In just three months, it increased conversion rate by 250%, tripled sales pipeline size, and saved 1320 hours.
Alipay has launched an invitation-based internal test for its AI version, which is the biggest overhaul since its launch 20 years ago. It core includes an "Assets" page and AI assistant "A-Bao", allowing users to let AI schedule various existing services through natural language, building an Agent service entrance.
OiiOii releases version 2.0, adding three major functions: intelligent canvas, video analysis replication, and Skill library, simplifying the AI video creation process, allowing ordinary people to make videos without complex operations. The platform currently launched a June discount event.
Anthropic releases an AI perception survey of 52,000 Americans, 71% of respondents support government regulation of AI, only 15% trust AI companies to self-regulate; 64% worry AI will take their jobs, 48% expect AI to tackle difficult diseases.
Researchers from Alibaba Qwen Large Model Application Team and other institutions proposed the GAP method, which solves the feature space mismatch problem in visual latent reasoning through three-layer alignment, and improves both average visual perception and multimodal reasoning performance on Qwen2.5-VL 7B.
Indian AI phone assistant Equal AI has completed a $30 million Series B funding round, bringing total accumulated funding to over $42 million. It currently has more than 1 million monthly active users, and can answer and screen calls, organize call content for users.
A new AI model named Le Chaton Fat allegedly developed by France's Mistral has gone viral in the AI community, with specs of 30T MoE and 1M context window, outperforming existing top models. It was later confirmed to be a hoax created by internet users.
Anthropic has withdrawn its ban on third-party applications programmatically calling Claude using subscription quotas, allowing third-party apps to deduct usage from subscription quotas normally. On the same day, a US user sued Anthropic for false advertising about usage quotas of its Max plans, seeking compensation.
Lin Junyang, former technical head of Qwen at Alibaba, has established a new AI Lab and completed the first round of financing. Tencent invested 20 million USD (approximately 130 million RMB), with a post-investment valuation of about 2 billion USD (approximately 13.5 billion RMB) for this round. Lin has already started seeking the next round of financing after completion.
MiniMax open-sourced its native multimodal flagship model M3 last Friday, with 428B total parameters and 23B activated parameters, and released MSA technical paper. It currently ranks first in global open-source model comprehensive intelligence ranking, with output speed increased to around 80 TPS.
Anker Innovations founder Yang Meng stated in an interview that smart hardware in the AI era should focus on three modules: perception, planning and control. After nearly three years of development, Anker launched noise-canceling headphones equipped with self-developed in-memory computing chips, whose call noise cancellation capability outperforms AirPods. Anker is now actively recruiting talents to promote AI transformation.
Peking University's Narwhal-Lab has open-sourced the Narwhal AI Code Risks project, which organizes risk cases, signals and paths of AI-generated code to help developers identify hidden dangers in advance and avoid losses.
Tsinghua University team launches AgentSociety², an integrated research environment for executable social science, enabling collaborative reproducible social science experiments between AI Social Scientists and Silicon Participants, providing new infrastructure for computational social science.
Former Gionee Global Vice President Yu Lei leads a nearly 60-person team to found Eclix, developing an AI-native smartphone without apps, adopting a dual-brain architecture, expected to launch this fall, targeting the premium mid-range market, with a limited first batch of 50,000 to 100,000 units.
Recruitment expert seb analyzed 1,680 LinkedIn resumes of Anthropic engineering employees and found that the company prefers infrastructure engineers with an average of 12 years of experience, Google is the largest previous employer source, and only 13.7% of employees hold PhD degrees.
Tencent Robotics X, in collaboration with Futian Laboratory and Hunyuan Team, open-sources the embodied intelligence model HyVLA-0.5. Built on sub-millimeter UMI with over 10,000 hours of collected data, it achieves over 90% success rate on RoboTwin 2.0 benchmark, and the success rate on real robot tasks approaches 100% after reinforcement training.
A developer added their open-source AI video editing Skill from GitHub to 360 Security Lobster Cloud version, making it a directly usable Video Highlight Expert Lobster that automatically edits videos and extracts highlights. The product also updated Skillify and other features this time.
In 2026, many Chinese universities set proportion limits on AIGC content in graduation theses. The detection is inaccurate, many graduates spend money on repeated testing and AIGC rate reduction, devoting lots of time and energy to prove they are human.
French AI startup Mistral AI is in talks to raise $3.5 billion, with a valuation of about $23.4 billion. It focuses on industrial custom AI and security models, providing AI services for European industrial enterprises and governments.
Fudan University and Tencent Hunyuan team propose Baton, pioneering a semantic blueprint guidance mechanism to solve audio-video out-of-sync problem in complex scenarios. On the Sem100 test set, prompt following accuracy improves by 32%, and audio-video desync metric improves by 30%.
Anthropic has launched Claude Design, a product that can generate high-precision interactive App prototypes based on descriptions, with outputs as code ready for direct development. Currently only the Claude Opus 4.8 model can achieve this functionality. OpenAI Codex has not launched a similar product yet due to insufficient capability of GPT-5.5.
Moore Threads open-sourced the MusaCoder-27B large model, which ranked first on the KernelBench benchmark with an 88.6% pass rate, outperforming models including Claude Opus. It can automatically generate GPU kernels, lowering development threshold, and the entire training process is completed on domestic GPUs.
OpenAI Codex launches self-goal generation feature. Humans only need to provide high-level intentions, and the model can independently set goals, split tasks, and schedule multiple agents to complete development. A developer has tested it, completing 14 functions independently in 18 hours at a cost of about $4.20.
Researchers define a new task of Knowledge-Intensive Video Generation (KIVI), construct KIVI-Bench benchmark with 1080 prompts, evaluate 7 mainstream models, find three types of systematic errors, current models still have gaps to practical use.
Meitu Imaging Research Institute (MT Lab) and University of Texas at Austin propose PE-Field framework, which extends traditional 2D positional encoding to structured 3D field, enabling DiT to directly process 3D spatial geometric information. The work has been accepted by ICLR 2026.
Anthropic researcher Vivek Nair published a long thread on 𝕏 sharing methodology for becoming an excellent AI researcher, pointing out that research capability can be obtained through deliberate practice, former Qwen head Junyang Lin retweeted and recommended it.