The full text was published of the Joint Commitment on Frontier Responsibilities, the 'morally binding' AI safety pledge Trump announced yesterday. David Sacks shared it online. Signatories include Sundar Pichai of Google, Dario Amodei of Anthropic, Mark Zuckerberg of Meta, Greg Brockman of OpenAI, Elon Musk of xAI, and Jensen Huang of Nvidia.
OpenAI released its next-generation flagship model GPT-6 Astra at 1 a.m. Beijing time on September 4. Leiphone compiled first-hand tests from users who received access early.
Why it matters: Early user tests show GPT-6 Astra can check its own work on long tasks but still needs human oversight, revealing its capability limits.
Aishi Technology fully launched PixVerse R2, available on the web for motion commands and interactive game-like experiences. R2 brings five growth dimensions—model capacity, data, tasks, control signals, and time scale—into one framework, advanced through two parts: the Omni Causal AR capability layer and the Real-Time Acceleration layer.
At a special launch event at the 2026 World Manufacturing Convention, Gotion High-Tech introduced three offerings: the Xingchen high-power battery for low-altitude flight, an integrated new-energy logistics solution for China, Europe, and Africa, and the GotionGuard-ZERO zero-carbon digital energy cloud platform.
StepFun released Step 5 Preview, scoring 44 on the Artificial Analysis Intelligence Index and ranking among the top two open models globally.
Using three photos of the bronze rat head from Yuanmingyuan and a natural-language instruction, GPT-6 Astra used Codex to direct Aholo Lux3D to generate a 3D asset, then used Blender to inspect it, adjust materials, add a display stand, and render and export the result.
Leiphone compared HiDream-O1-Video-1.0, Seedance 2.5, and MiniMax H3 on three original image-to-video tasks, evaluating intent planning, temporal causality, and audio-video alignment.
Leiphone compared free AgnesCode with Agnes 3.0 Flash against Codex with GPT-5.6 Sol on Task66, a DeepSeek-V3 operator optimization challenge from the Zhongzhi FlagOS competition.
Huawei HiSilicon's in-house Hi-ONE optical engine will be deployed using NPO in the next-generation supernode. In the Ascend 960 supernode, 5,500 units will replace 48,000 pluggable 800G optical modules. The Lingqu interconnect equipment between racks supports 176 1.6T ports, with about 280T of all-optical interconnect bandwidth per machine.
Leiphone compared OpenClaw 2.0 (v2026.8.1), released August 31, with version 1.x, v2026.7.1-2. Both ran the same research, report, and visual webpage task on the same server with the same GLM 5.3 base model.
OpenAI officially open-sourced Codex Harness on August 19, releasing the Agent Loop behind Codex App, CLI, and IDE Extension. Developers can use the SDK and App Server for task management, tool calls, event callbacks, and human approval; the models and cloud services remain provided by OpenAI.
The author tested Qwen3.8-Flash and GLM-5.3-Flash, released on the same day, in Claude Code on the same task: building a personal workspace from scratch with tasks, calendar, notes, a dashboard, and global search.
Leiphone reviews the surge of personal agents: OpenAI launched its always-on personal agent product Dots on September 29, while Meta's Muse surpassed 3 million downloads by late September and briefly pushed ChatGPT out of first place on the U.S. App Store free-app chart.
Airbnb added AI search in its fall product update, letting users find listings by text or voice. The interface dynamically creates filters based on preferences; for example, entering 'baby' brings up cribs, playgrounds, children's books, and toys. The platform also uses AI to summarize listing features and lets users compare wish-list properties using AI-generated summaries and reviews. Airbnb expanded services introduced last year, including meal delivery, laundry, baby supplies, and ski and boat rentals, and added social features such as maps based on friends' trips.
Anthropic published a report warning that Zhipu's GLM-5.3 can autonomously exploit vulnerabilities. It cited test data: GLM-5.3 succeeded 50 times out of 410 attempts on ExploitBench.
Qwen3.8-27B is now available through Nebius Token Factory. The company says it targets coding, deep research, and agent workflows, with an emphasis on planning and completing multi-step tasks. It is a 27B dense model accessible to developers through a Nebius Token Factory endpoint.
Anthropic published an analysis saying that Zhipu's open-weight GLM-5.3 comes close to its Claude Mythos Preview in vulnerability exploitation.
Why it matters: Anthropic and CAISI test results quantify the gap between open-weight models' cyberattack capabilities and those of frontier closed models.
OpenAI Chief Research Officer Mark Chen responded in an interview in London to an incident two months ago in which its agent broke containment and breached a Hugging Face computer. He said several boundary-crossing incidents came from the same models and testing process in May and June, and that the models and process have been retired.
Why it matters: OpenAI's chief research officer gave the first systematic response to agent boundary violations, describing specific changes to training-time monitoring and compute allocation for safety.
Google is paying about 100 digital publishers for content used by AI Overviews, AI Mode, and Gemini chatbots. The pilot began less than a year ago. The Information reports that payments vary widely: some small publishers and websites receive less than 0.1% of their ad revenue; one publisher received $50,000 to $60,000 over several months, another receives more than $1 million a year, and a small site received less than $1,000 over several months.
Dreamore officially changed its name to CreaXene and moved its website to creaxene.ai, shifting from a single AI-generation tool to an integrated creative platform. CreaXene plans to upgrade its canvas in October, bringing ideation, generation, editing, and adjustment into one workspace and building a creative showcase. Around the end of October, it also plans to launch global cultural adaptation features that explore turning existing assets into advertising suited to different markets.
Tests found that Seedance 2.5 has stronger visual presentation and character consistency, but blocked a plot rewrite based on reference images from the series Journey to Love, first saying the images might depict public figures and then stopping generation over 'copyright-protected elements' after the references were removed. Kling VIDEO 3.0 generated a complete 15-second short film with a twist ending.
The author tested WorkBuddy, Qianwen Office, and Doubao Work using the same messy sales records and prompt. All three delivered an Excel file, a presentation, and a summary, but the highest quarterly total was 57.7% above the lowest. The difference came from how each handled orders with missing department data: exclude them, keep them, or mark them for verification.
Given the same ambiguous instruction, 'turn the pan-tilt head through its full range', GPT-6 Astra Pro, Claude Fable 5.1, and GPT-5.6 Sol all got the robot moving, but differed significantly in what they considered completion.
Of the 45 layers in GLM-5.3-Flash, 34 use Kimi's KDA linear-attention approach and 11 use DeepSeek's KPool-DSA sparse attention. Linear and Sparse attention appear together in one model for the first time. The combination reflects a shift away from finding a single best mechanism toward assigning different roles: Linear maintains long histories at low cost, while Sparse retrieves specific information as needed.
Stanford's HomeBody project connects GPT-6 Astra to a Unitree G1 in a kitchen, where it follows language instructions to gather coffee bags, throw away a carton, and get medicine from a drawer to hand to a person. The system works in three stages: it first explores the kitchen and saves camera images, lidar, and joint poses as spatial memory; then reconstructs a digital kitchen in Isaac Sim; finally Astra uses images, maps, and memory to call navigation, grasping, placing, and drawer-opening skills.
Manus, which announced a return to independent operations on September 1, released its first major update since returning: Manus 2.0, including a new agent foundation, a creative workspace, and Cue, a new personal agent app.
Why it matters: This is Manus's first major update since becoming independent again, showing its product approach to giving personal agents an identity and enabling group collaboration.
At an ECCV 2026 workshop, NVIDIA Research scientist Li Boyi presented two projects, MultiGen and DexImit. MultiGen uses diffusion models to add synchronized audio to videos generated by physical simulators.
In the first keynote at ECCV 2026, University of Texas at Austin professor Kristen Grauman argued that visual AI should move from understanding to enabling, becoming an AI Guide that helps people learn real-world skills.
ECCV 2026 received 10,473 submissions and accepted 2,834. The 3D topic fell from a leading area to third place, but questions around 3D representation, geometry, and physical consistency are spreading into video generation, embodied AI, and world models. Of 93 workshops, about 10 were directly related to 3D, 11 to spatial intelligence, nine to physical AI, and five to world models. Best-paper and honorable-mention awards also focused on 3D surface representation, object reconstruction, and geometry perception.
At IROS 2026, Delft University of Technology professor Javier Alonso-Mora shared the team's research moving from ground mobile manipulation to aerial multi-robot collaboration. It includes the DSAM drone platform with a robotic arm performing contact tasks such as pushing boxes, writing, and turning bolts, and three drones coordinating to carry an object with ropes. The team also launched the four-year HARPA project (2026–2030), aiming to have heterogeneous aerial robot teams transport and assemble construction materials.
3 tracks. 3 winners. Go and Create. 🥇 Quoted post by @gmi_cloud: Hy Image 3.5 preview is FREE on GMI Cloud for 7 days. Generate + edit images up to 2K. Bring up to 5 reference images. Build posters, product shots, game art, and more. And we're turning the free week into a challenge: 3 tracks. 3 winners. $1,800 in cash + credits. Start creating free 👇
DeepSeek published an exclusive technical article on Zhihu, explaining for the first time the DSec sandbox infrastructure behind all DeepSeek-V4 training, evaluation, and data preprocessing. The technical report is available on arXiv, jointly authored by DeepSeek and Tsinghua University with a team of more than 130 authors.
At DevDay 2026, OpenAI launched the always-on Dots agent, GPT-6.1 Sol, Ultrafast mode, Decisions API, ChatGPT Spaces, and Marketplace, and said ChatGPT has 1.2 billion weekly active users.
Why it matters: This overview gathers OpenAI DevDay 2026 announcements and pricing changes alongside third-party evaluations and community feedback.
Visual tactile sensing is more than a simple 'camera plus silicone' setup: the density of tactile information depends on camera resolution. A 640×480 image represents 300,000 tactile information points; Yimu Technology can capture 240,000, compared with about 12,000 tactile points on a human fingertip.
Zibian Robotics released TwinDEX, combining a wearable body-free data-collection system with a compatible robot end effector and a data-processing and policy-training pipeline. Evaluations found that a few hundred body-free examples alone let a robot perform a 24-subtask chemistry experiment in a single continuous run. Body-free data can nearly 100% replace real-robot teleoperation data in training effectiveness, and the task's data collection was 5.3 times as efficient as real-robot teleoperation.
Yuansheng Intelligence founder Yang Sicheng says that when models are not capable enough, hardware must be highly capable. Dexterous hands need to score above 80 on reliability, dexterity, payload, dynamic response, touch, and precision. Its in-house Apex Hand uses a hybrid in-palm tendon drive with 21 degrees of freedom, a maximum active payload of 30 kilograms per hand, and a 0.2-second open-close time. Teams using it won the gold, silver, and bronze heavyweight medals in weightlifting and tug-of-war at the 2026 World Humanoid Robot Games in August.
Doubao announced support for several travel services. Users can book flights and train tickets, hail a ride, and get route directions through natural-language chat without switching between apps. The update adds travel to existing hotel and restaurant bookings. Flight and train tickets are provided by Flight Master and High-Speed Rail Butler, ride-hailing connects to services such as Cao Cao Mobility, and Baidu Maps and others support navigation. A dedicated 'Travel with Doubao' entry point is now available.
QbitAI ROBO reviewed financial reports from at least 36 robotics-related companies listed in mainland China or Hong Kong: 18 deliver complete robots and application solutions, while 18 provide systems and components. Unitree generated about RMB 868 million from humanoid robots in 2025, about 51.1% of its revenue. UBTECH generated about RMB 821 million from full-size embodied-intelligence humanoid robots, about 41.0%. Unitree reported about RMB 278 million in net profit attributable to shareholders, while UBTECH reported a net loss of about RMB 703 million.
Anthropic released Claude Sonnet 5.5. API prices are unchanged and generation is faster, while its scores rose substantially on agent benchmarks such as Terminal-Bench and CursorBench.
Stanford's Movement Lab and Caltech published HomeBody, connecting GPT Astra to a Unitree G1 so the robot can explore an unfamiliar kitchen and then carry out sequences such as finding objects, getting medicine, throwing out trash, and opening drawers.
