## 🔍 Core Insights **The AI model ecosystem** is shifting from single-point competition toward "free combination": **Doubao** has integrated third-party models such as **Kimi** and **DeepSeek**, **Baidu** has launched the **Token Exchange** trading platform, and the power to choose model-as-a-service is accelerating toward users [0][2]. Meanwhile, **Runway** has open-sourced the robotics model **Praxis-1** and **Volcano Engine** has released **Seedance 2.5 flash**, as the cost-performance race in multimodal and embodied intelligence heats up in parallel [1][6]. On the regulatory side, the U.S. Department of Commerce plans to tighten the subleasing pathways for **Nvidia Blackwell** chips, with geopolitical maneuvering continuing to affect compute supply [14]. ## 🚀 Key Updates - **Doubao Work integrates multiple third-party models including Kimi and DeepSeek** [0]: On the same platform, users can autonomously select models based on the task and adjust how quota consumption is displayed. - **Runway open-sources robotics model Praxis-1** [1]: Open weights under Apache 2.0, 15.6x faster than the GPT-6 Astra robotic arm. - **Baidu launches Token Exchange, a large-model trading platform** [2]: Users can compare models, pricing, and performance; qualifications are verified before listing, and model-switching mislabeling is detected. - **Volcano Engine releases Seedance 2.5 flash video model** [6]: Generation speed doubled, 720P priced at about 0.99 yuan/second, a reduction of about 34%. - **Tongxin Intelligence completes funding of hundreds of millions of yuan** [12]: Its first AGI inference chip is planned for prototype validation in December, with funds used for pre-tapeout architecture design. - **New U.S. Department of Commerce rules would restrict Chinese AI companies from renting Nvidia chips** [14]: The rules would require export licenses from seven countries, with buyers required to disclose recipients and use cases and accept monitoring. - **Google releases Gemini Decision 1 decision model** [15]: Completes closed-set decisions in a single forward pass, supports 256K context, input at $0.10/1M tokens. - **MIT trains climate simulator with synthetic data** [18]: Running a full Earth system model just once is enough to generate optimized data, greatly shortening production time. ## 🔗 Sources [0] Just now, Doubao Work officially announced integration with Kimi and DeepSeek, AI model freedom has arrived — https://www.bestblogs.dev/article/61ee281a2d?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item [1] Runway open-sources robotics model Praxis-1, 15.6x faster than the GPT-6 Astra robotic arm — https://aihot.news/items/ifml1k624yaj9mf80iu2c8k2r [2] Baidu launches Token Exchange large-model trading platform — https://aihot.news/items/om2hap22jk1qapfbxl4410ro7 [6] ByteDance's Volcano Engine releases Seedance 2.5 flash video model: speed doubled, price reduced by about 34% — https://aihot.news/items/u3a5s72ayl5qo1lbqe2x6ggua [12] Tongxin Intelligence completes funding of hundreds of millions of yuan, first AGI chip planned for prototype validation in December — https://aihot.news/items/t5xynkv3vqe2j6wf9aeb2v6t7 [14] The Information: New U.S. Department of Commerce rules would require export licenses from seven countries, restricting Chinese AI companies from renting Nvidia Blackwell and other chips — https://aihot.news/items/q