Author: RadarAI Editorial
Editor: RadarAI Editorial
Last updated: 2026-07-20
Review status: Editorial review pending
Brief
速报
官方
AI动态
开源
Edge AI and embodied intelligence are rapidly transitioning from lab research to mass production. Baichuan Intelligence's MiniCPM-Robot series achieves state-of-the-art open-source local navigation performance at just 1.5B parameters; SenseTime has turned its domestic AI compute business profitable via heterogeneous hybrid inference, processing over 10 trillion tokens per day [7][24]; meanwhile, Qwen 3.8-Max-Preview sets a new open-source large model scale record with 2.4 trillion parameters [15].
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Key Insights
**Edge AI** and **embodied intelligence** are accelerating from lab research toward mass production. **Baichuan Intelligence**'s MiniCPM-Robot series achieves the best-in-class open-source local navigation performance at just 1.5B parameters; **SenseTime**, leveraging heterogeneous hybrid inference, has turned its domestic AI compute business profitable—processing over **10 trillion tokens per day** [7][24]. Meanwhile, **Qwen 3.8-Max-Preview**, with **2.4 trillion parameters**, sets a new record for open-source large model scale [15].
## 🚀 Key Updates
- **Baichuan Intelligence launches the MiniCPM-Robot series of embodied models** [3]: Open-sourced a 1.5B-parameter vision-language-action (VLA) model at WAIC, resolving long-horizon memory challenges and achieving best-in-class open-source local tracking and navigation.
- **Baichuan Intelligence releases MiniCPM-RobotManip/Track dual models** [4]: Integrating memory optimization and the PhyAI inference framework, enabling efficient on-device deployment and strengthening its foundational capabilities in embodied intelligence.
- **SenseTime achieves positive gross margin on domestic AI compute** [24]: Heterogeneous hybrid inference + end-to-end autonomous scheduling boosts token throughput by 25×, turning gross margin positive.
- **Qwen 3.8-Max-Preview (2.4T parameters) weights released** [15]: Alibaba Cloud announces imminent open-sourcing; users can preview early access via Token Plan and Qoder platforms.
- **PPIO builds an 'AI Token Factory' for Agents** [5]: Launches an intelligent model gateway and Agent sandbox—delivering ~20% performance uplift and 50–60% cost reduction.
- **Riemann-1.0 learns embodied skills from 200,000 hours of human video** [17]: Achieves 62.6% SOTA on the RoboCasa-365 household task benchmark, validating the generalization value of first-person human video data.
- **Perplexity AI open-sources Wandr—a deep-research benchmark** [18]: Designed specifically to evaluate AI agents' real-world capability in broad, in-depth research-oriented information work.
- **WAIC reaches new consensus on the 'Robot ChatGPT Moment'** [8]: Industry experts jointly project this inflection point may arrive within **2–5 years**, contingent upon breakthroughs across three technical frontiers: data, representation, and closed-loop control.
## 🔗 Sources
[1] Why Did Baichuan Intelligence Light the Torch for Edge AI? — https://www.bestblogs.dev/article/6410b00621?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[2] Kimi K3's KDA Mechanism Improves Attention Efficiency—but Requires *More* GPU, HBM, DRAM, and Network Bandwidth, Not Less! — https://www.bestblogs.dev/article/9f436bf6d7?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[3] Competitors All Guessed Wrong—It Scored 53! Baichuan's 1.5B Model Stuns at WAIC, Halving Costs and Shattering the Ceiling for Domestic Robots — https://www.bestblogs.dev/article/36202ebcb6?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[4] Baichuan Intelligence Enters Embodied Intelligence—with Its Full Stack — https://www.bestblogs.dev/article/31aefe26d7?utm_source=rss&utm_medium=feed&utm
Edge AI and embodied intelligence are accelerating from lab research toward mass production. Baichuan Intelligence's MiniCPM-Robot series achieves the best-in-class open-source local navigation performance at just 1.5B parameters; SenseTime, leveraging heterogeneous hybrid inference, has turned its domestic AI compute business profitable—processing over 10 trillion tokens per day [7][24]. Meanwhile, Qwen 3.8-Max-Preview, with 2.4 trillion parameters, sets a new record for open-source large model scale [15].
🚀 Key Updates
- Baichuan Intelligence launches the MiniCPM-Robot series of embodied models [3]: Open-sourced a 1.5B-parameter vision-language-action (VLA) model at WAIC, resolving long-horizon memory challenges and achieving best-in-class open-source local tracking and navigation.
- Baichuan Intelligence releases MiniCPM-RobotManip/Track dual models [4]: Integrating memory optimization and the PhyAI inference framework, enabling efficient on-device deployment and strengthening its foundational capabilities in embodied intelligence.
- SenseTime achieves positive gross margin on domestic AI compute [24]: Heterogeneous hybrid inference + end-to-end autonomous scheduling boosts token throughput by 25×, turning gross margin positive.
- Qwen 3.8-Max-Preview (2.4T parameters) weights released [15]: Alibaba Cloud announces imminent open-sourcing; users can preview early access via Token Plan and Qoder platforms.
- PPIO builds an 'AI Token Factory' for Agents [5]: Launches an intelligent model gateway and Agent sandbox—delivering ~20% performance uplift and 50–60% cost reduction.
- Riemann-1.0 learns embodied skills from 200,000 hours of human video [17]: Achieves 62.6% SOTA on the RoboCasa-365 household task benchmark, validating the generalization value of first-person human video data.
- Perplexity AI open-sources Wandr—a deep-research benchmark [18]: Designed specifically to evaluate AI agents' real-world capability in broad, in-depth research-oriented information work.
- WAIC reaches new consensus on the 'Robot ChatGPT Moment' [8]: Industry experts jointly project this inflection point may arrive within 2–5 years, contingent upon breakthroughs across three technical frontiers: data, representation, and closed-loop control.
🔗 Sources
[1] Why Did Baichuan Intelligence Light the Torch for Edge AI? — https://www.bestblogs.dev/article/6410b00621?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[2] Kimi K3's KDA Mechanism Improves Attention Efficiency—but Requires More GPU, HBM, DRAM, and Network Bandwidth, Not Less! — https://www.bestblogs.dev/article/9f436bf6d7?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[3] Competitors All Guessed Wrong—It Scored 53! Baichuan's 1.5B Model Stuns at WAIC, Halving Costs and Shattering the Ceiling for Domestic Robots — https://www.bestblogs.dev/article/36202ebcb6?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[4] Baichuan Intelligence Enters Embodied Intelligence—with Its Full Stack — https://www.bestblogs.dev/article/31aefe26d7?utm_source=rss&utm_medium=feed&utm
← Back to Updates