Author: RadarAI Editorial
Editor: RadarAI Editorial
Last updated: 2026-06-01
Review status: Editorial review pending
Brief
速报
官方
AI动态
开源
Embodied world models are experiencing an explosive wave of open-source progress, with τ0-WM and STI-WM recently released—marking a new phase for robots' 'slow thinking' decision-making and the real-world deployment of physical AI. Zhipu AI has surged to become the world's highest-valued open-source software company through its full-model open-source strategy, with a valuation exceeding Xiaomi's [1]. Meanwhile, Anthropic is reportedly suspected of deliberately degrading older model performance—a revelation sparking deep ethical concerns regarding commercial practices among large-model vendors [10].
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Key Insights
**Embodied world models** are witnessing explosive open-source progress: τ0-WM and STI-WM have been successively released, signaling a new stage in robots' 'slow thinking' decision-making and the real-world deployment of physical AI. **Zhipu AI**, leveraging its full-model open-source strategy, has risen to become the **world's highest-valued open-source software company**, with a valuation surpassing Xiaomi's [1]. At the same time, **Anthropic** is alleged to have deliberately downgraded older model performance—an incident triggering profound ethical scrutiny of large-model vendors' commercial conduct [10].
## 🚀 Major Updates
- **World's largest open-source pre-trained embodied world model τ0-WM launched** [4]: Trained on 17,800 hours of real-robot teleoperation data; enables test-time computation for robots to 'think thrice before acting'.
- **Fudan University's MoShen Intelligence unveils STI-WM—the Spatio-Temporal Integrated World Model** [9]: A natively embodied 'brain' designed specifically for robots, emphasizing spatio-temporal unified modeling and physics-consistency constraints.
- **Zhipu AI becomes the world's highest-valued open-source software company** [1]: Its market capitalization now exceeds Xiaomi's—and equals more than two JD.com's—thanks to its sustained open-sourcing of all large models.
- **Microsoft open-sources SkillOpt framework** [7]: First to treat agent skill documentation as trainable parameters, enabling self-evolution of skill docs; achieves state-of-the-art performance across all 52 benchmark evaluations.
- **Renmin University's Gaoling Institute releases first Rubrics survey paper** [8]: Systematically constructs a unified, multi-dimensional evaluation framework for defining 'good answers' in the Agent era—covering definition, construction, training, and evaluation across the full lifecycle.
- **AI age verification defeated by stick-figure drawings** [6]: Users bypassed age checks on platforms like Discord using fake beards and finger-drawn doodles; Meta responds with new defenses, including skeletal-structure scanning.
- **Anthropic accused of intentionally degrading legacy model performance** [10]: Media reports allege deliberate performance reduction of older models ahead of Opus 4.8's release—triggering a crisis of trust in model quality.
- **General-purpose Agents will replace operating systems** [11]: Industry consensus holds that apps will either vanish or evolve into CLI/MCP plugins; SaaS providers must offer both CLI interfaces and skills to integrate into the Agent ecosystem.
## 🔗 Sources
[1] Zhipu AI's market cap has surpassed Xiaomi's, making it the world's highest-valued open-source software company — https://www.bestblogs.dev/status/2061009832963154240?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[4] τ0-WM: The largest-scale pre-trained open-source embodied world model arrives — https://www.bestblogs.dev/article/22e0cf5c?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[6] A primary-school student drew a fake beard and fooled AI age verification—Silicon Valley engineers fell silent — https://www.bestblogs.dev/article/0d041316?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[7] 3,300 GitHub stars in one week: Microsoft kicks off skill self-evolution! Training skill documentation just like neural networks — https://www.bestblogs.dev/article/7b8c6b8d?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[8] Rubrics Survey: In the Agent Era, How Do We Define a 'Good Answer'? — https://www.bestblogs.dev/article/d170c0c0?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[9] Fudan's 'Iron Triangle': Pioneering the Optimal Path for Physical AI
Embodied world models are witnessing explosive open-source progress: τ0-WM and STI-WM have been successively released, signaling a new stage in robots' 'slow thinking' decision-making and the real-world deployment of physical AI. Zhipu AI, leveraging its full-model open-source strategy, has risen to become the world's highest-valued open-source software company, with a valuation surpassing Xiaomi's [1]. At the same time, Anthropic is alleged to have deliberately downgraded older model performance—an incident triggering profound ethical scrutiny of large-model vendors' commercial conduct [10].
🚀 Major Updates
- World's largest open-source pre-trained embodied world model τ0-WM launched [4]: Trained on 17,800 hours of real-robot teleoperation data; enables test-time computation for robots to 'think thrice before acting'.
- Fudan University's MoShen Intelligence unveils STI-WM—the Spatio-Temporal Integrated World Model [9]: A natively embodied 'brain' designed specifically for robots, emphasizing spatio-temporal unified modeling and physics-consistency constraints.
- Zhipu AI becomes the world's highest-valued open-source software company [1]: Its market capitalization now exceeds Xiaomi's—and equals more than two JD.com's—thanks to its sustained open-sourcing of all large models.
- Microsoft open-sources SkillOpt framework [7]: First to treat agent skill documentation as trainable parameters, enabling self-evolution of skill docs; achieves state-of-the-art performance across all 52 benchmark evaluations.
- Renmin University's Gaoling Institute releases first Rubrics survey paper [8]: Systematically constructs a unified, multi-dimensional evaluation framework for defining 'good answers' in the Agent era—covering definition, construction, training, and evaluation across the full lifecycle.
- AI age verification defeated by stick-figure drawings [6]: Users bypassed age checks on platforms like Discord using fake beards and finger-drawn doodles; Meta responds with new defenses, including skeletal-structure scanning.
- Anthropic accused of intentionally degrading legacy model performance [10]: Media reports allege deliberate performance reduction of older models ahead of Opus 4.8's release—triggering a crisis of trust in model quality.
- General-purpose Agents will replace operating systems [11]: Industry consensus holds that apps will either vanish or evolve into CLI/MCP plugins; SaaS providers must offer both CLI interfaces and skills to integrate into the Agent ecosystem.
🔗 Sources
[1] Zhipu AI's market cap has surpassed Xiaomi's, making it the world's highest-valued open-source software company — https://www.bestblogs.dev/status/2061009832963154240?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[4] τ0-WM: The largest-scale pre-trained open-source embodied world model arrives — https://www.bestblogs.dev/article/22e0cf5c?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[6] A primary-school student drew a fake beard and fooled AI age verification—Silicon Valley engineers fell silent — https://www.bestblogs.dev/article/0d041316?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[7] 3,300 GitHub stars in one week: Microsoft kicks off skill self-evolution! Training skill documentation just like neural networks — https://www.bestblogs.dev/article/7b8c6b8d?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[8] Rubrics Survey: In the Agent Era, How Do We Define a 'Good Answer'? — https://www.bestblogs.dev/article/d170c0c0?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[9] Fudan's 'Iron Triangle': Pioneering the Optimal Path for Physical AI
← Back to Updates