Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

China's PhysBrain 1.5 tops the global open-source ranking for physical AI

PhysBrain 1.5, a physics-focused AI model from a Chinese team, has reached the top of a global open-source leaderboard, according to QbitAI, which frames the result as clearing the hardest stretch of the physical closed loop. The report puts its spatial intelligence on par with GPT-6 Astra.

Published

PhysBrain 1.5, a physics-focused AI model from a Chinese team, has reached the top of a global open-source leaderboard, according to a report from QbitAI. The outlet frames the result as clearing the hardest stretch of the physical closed loop. Physical AI refers to models that understand and act on the physics and spatial relationships of the real world. Compared with text-only systems, it demands stronger spatial perception, motion prediction, and closed-loop interaction. The report says PhysBrain 1.5's spatial intelligence is on par with GPT-6 Astra, and that this strength carried it to first place on the open-source ranking. Leaderboard position matters because it makes results comparable. A top ranking implies the model can be evaluated against peers on the same spatial-reasoning and physics-prediction tasks, rather than only in vendor-selected demos. Physics capability is widely treated as a foundation for robotics, autonomous driving, and industrial simulation. When a model makes progress in predicting physical behavior, the distance from release to real deployment tends to shorten. That is why the result drew attention: it concerns not just a ranking slot but whether one of the hardest loops in the physical world has actually moved forward. What to watch next: whether the ranking holds up under independent evaluation, how much of the benchmark methodology is published, and how the model performs in real robot or simulation environments.

Why it matters

If the spatial-intelligence lead can be reproduced under independent evaluation, physical AI should reach robotics and autonomous-driving scenarios faster, giving Chinese teams a stronger open-source position in the field.

Physical AI开源模型空间智能
Back to realtime news

Nearby Updates

All

09/14, 11:01

DeepSeek-V4.1-Flash: 552B-Parameter MoE Built for Efficient Inference

A technical breakdown of the DeepSeek-V4.1-Flash model card describes a multimodal mixture-of-experts model with a 552B-parameter backbone that activates only about 8B parameters at a time. The weights ship under an MIT license with a 1M-token context window, but the model still trails the larger V4-Pro family on several reasoning benchmarks.

09/14, 11:55

Chinese Models Lead Weekly Token Volume for 20th Straight Week as DeepSeek V4.1 Flash Hits No. 6

National Business Daily, using the latest OpenRouter data, calculates that global large-model token consumption reached 127 trillion tokens in the week of Sept. 7 to Sept. 13, with Chinese models at 61.17 trillion tokens, leading for a 20th consecutive week. DeepSeek V4.1 Flash, released Sept. 10, climbed to sixth place within three days, while Chinese models took four of the top five slots.

09/14, 12:05

OpenAI Pitches AI-Native ChatGPT Ads That Open a Brand Chat, Not a Website

Digiday reports that OpenAI has introduced a new AI-native ad format to select clients, attaching a branded business agent to the ad so that a click opens a chat inside ChatGPT instead of sending users to the advertiser's site. Wayfair is trialing the format, and OpenAI CFO Sarah Friar has described today's response ads as only a "basic starting point."

09/14, 09:51

DeepSeek Routes All V4-Pro API Traffic to V4.1-Flash at Flash Rates

DeepSeek now routes all API traffic aimed at V4-Pro to V4.1-Flash and bills it at Flash rates, according to Pandaily. Requests that need V4-Pro capability are therefore answered by V4.1-Flash, changing the cost picture for callers.