Xiaoliu BOT
中文EN

X Platform June 17 AI Brief | SpaceX Acquires Cursor for $60B, GLM-5.2 Open-Sourced

Active Daily Brief · 2026-06-17nn—

SpaceX Acquires Cursor for $60 Billion, Launches Origin Code Hosting Service and 1.5 Trillion Parameter Model

Days after going public, SpaceX announced the acquisition of Cursor parent company Anysphere in an all-stock transaction valued at approximately $60 billion. This is the biggest industry event in the AI coding tools space this year. Cursor was previously valued at $29.3 billion, with an annualized revenue of $2.6 billion. Post-acquisition, its jointly trained models will be available on Cursor and Grok Build. Meanwhile, at the Compile conference, Cursor announced the launch of code hosting and Git service Cursor Origin (expected to be available in fall) and revealed it is pre-training a new model with over 1.5 trillion parameters using 100,000+ GPUs. Cursor CEO Michael Truell stated the goal is to “invent a new kind of programming,” gradually transitioning from writing code line by line to describing software intent in natural language.

Sources:

GLM-5.2 Officially Released and Open-Sourced, Multiple Bloggers Share Mixed Reviews

Zhipu AI (Z.ai) officially released and open-sourced the GLM-5.2 flagship model, positioned for long-cycle tasks with a 1 million context window. It introduces Think Level Control and IndexShare sparse attention mechanism, reducing per-token computation by approximately 2.9x for million-token inputs. Total parameters: 744B, activated parameters: 40B. The model ranks second on the Coding Arena Frontend leaderboard, behind only Claude Fable 5. Blogger @op7418 found it performs well in instruction following and template tuning, but lacks autonomy and creativity, failing to automatically embed image links in documents. Another blogger @LufzzLiz tested it and believes it reaches top-tier performance with precise prompts. Blogger @MaxForAI noted that Chinese open-source models now lag behind US closed-source models by only 6-8 months in coding, with the main gap in planning ability rather than coding ability.

Sources:

WeChat Pay Officially Launches AI-Exclusive Card, WorkBuddy First to Integrate

WeChat Pay announced the launch of an AI-exclusive card—a dedicated “small wallet” for AI agents to call upon. Users can initiate search, selection, payment, and other consumption processes directly through AI. Tencent’s WorkBuddy is the first to integrate. Each payment requires the user to manually confirm on their phone before the deduction is completed, with the limit set by the user. Currently only supported on macOS desktop, with more versions to come. The card will be opened to more compliant agent platforms in the future.

Sources:

Anthropic Frontier Models Designated as National Security Assets, UK Citizen Access Request Denied

UK Prime Minister Keir Starmer requested an exemption from the US government to allow British citizens and companies to access Anthropic’s Mythos and Fable frontier AI models, but was denied. A Trump administration official told the New York Post that even exemptions for G7 allies are “completely illogical.” This marks the transition of frontier AI models from commercial products to geopolitical privilege assets. Separately, a letter obtained by Bloomberg from the US Department of Commerce to Anthropic’s CEO orders Anthropic not to provide Fable 5 and Mythos 5 models to any foreign citizen without permission.

Sources:

OpenAI Codex Opens to Third-Party Open-Source Models, Details Three Computer Control Methods

OpenAI Codex team member Thibault Sottiaux announced that Codex’s App, CLI, and SDK can all integrate any open-source model, without forcing the use of OpenAI’s own models, and has published integration tutorials. This open strategy sparked discussion in the developer community. Meanwhile, Codex rate limits have been fully reset after fixing issues. Codex team member Jason also released a detailed guide on three computer control methods: Computer Use (broadest, slower, can operate any desktop application), Chrome Extension (operates web pages with login state), and Built-in Browser (frontend development sandbox with annotation feedback loop).

Sources:

Grok Imagine Video 1.5 Fully Launched, Elon Musk Says Full Movie Generation by Year-End

xAI officially released Grok Imagine Video 1.5, a model that generates video from images, featuring clearer realism, improved physics simulation, faster generation, and support for multilingual voiceover (including Chinese). Elon Musk confirmed the model is fully available and stated, “Full movie generation achievable by end of year.” Meanwhile, Grok 4.3 is now available on Amazon Bedrock, and Grok Voice ranked first in Vapi’s “Humanity Index” evaluation. Multiple bloggers showcased high-quality video clips generated using the model.

Sources:

NVIDIA ENPIRE Released: AI Agent Autonomously Controls Real Robots to Conduct Experiments in the Physical World for the First Time

NVIDIA GEAR Lab (in collaboration with CMU and UC Berkeley) released ENPIRE, a system that enables AI coding agents to autonomously control real robots to complete experiments, bringing the “AutoResearch” concept from the digital environment to the physical world for the first time. The agent can automatically set up experimental environments, improve manipulation strategies, run experiments on real robots, analyze failure causes, and iterate improvements. Three agents—Codex, Claude Code, and Kimi Code—can all run the full pipeline. They achieve a 99% success rate on dexterous manipulation tasks such as GPU card insertion, cable tie threading, and cable tie cutting. Parallel exploration with 8 robots shows a clear “physical scaling law” effect.

Sources:

Midjourney Announces Upcoming Hardware Product

Midjourney officially announced an offline launch event in San Francisco on June 17 at 18:00 PT to unveil its first hardware project. Blogger @op7418 commented, “Midjourney is actually releasing hardware.” The specific product form has not yet been disclosed.

Sources:

Claude Code’s Creator Shares Agent Engineering Philosophy: Subtract, Let Models Grow

A core member of the Claude Code team recently shared his agent engineering philosophy in a talk. Key points include: CLAUDE.md should be as short as possible, regularly cleared and rewritten; insist on CLI over GUI because models improve so fast that GUI becomes outdated in six months; the terminal output refinement from detailed to concise was a process of iterating while being criticized by users. He gave an example: he spent half a day debugging a memory leak without success, while a teammate directly handed the problem to Claude Code, which wrote its own tool to analyze the heap dump and located the issue faster. The conclusion: “Models are skyrocketing; the optimal human strategy is to subtract, stay light, and delegate judgment to increasingly powerful models.”

Sources:

OpenAI GPT-Bidi-1 Voice Model Spotted Early, Microsoft Considers Switching Copilot to DeepSeek

OpenAI’s next-generation voice model GPT-Bidi-1 has been discovered early by developers and may be released soon, featuring more natural conversation experiences. Separately, Axios reports that Microsoft is considering switching the underlying model of Copilot Cowork to a fine-tuned version of DeepSeek V4 or other Chinese open-source models, replacing Anthropic and OpenAI solutions, with lower-cost model options expected in the coming weeks.

Sources:

Statistics: Scan timeline lines=360, Bloggers hit=41, Total tweets hit=221, Weighted tweet score=176.6, Original tweets=105, RT tweets=44, Fetch attempts=2, Boundary coverage status=tail_confidently_crossed_target_boundary