[Baidu Releases Wenxin ERNIE 5.5 Flash Preview: 11 Hours to Develop a Wuxia Game in C++] Beating AI News: Baidu has released Wenxin ERNIE 5.5 Flash Preview. This is the first post-trained version of the 5.5 Flash Base, which was open-sourced last week, focusing on improving programming, agent, and complex task execution capabilities. The model has a total of 280 billion parameters, with 14 billion parameters activated per cycle, and supports 1 million token contexts. Baidu showcased a 3D Wuxia game, "Yunxiu Xing," developed using the model. In an 11-hour task, the model completed the Unreal Engine 5.8 game development using C++ and integrated assets from different engine versions. Interestingly, Wenxin 5.5 Flash is a pure text model; no game visuals are displayed. It converted issues discovered in gameplay feedback, such as characters floating in mid-air, incorrect movement directions, and camera misalignment, into executable code for testing, and then repeatedly modified it based on the test results. Pre-existing asset packages were used for the scenes and characters. Baidu also conducted specialized training for professional software such as Blender and CAD. In the official demonstration, two agents worked together to create the airport and aircraft, ultimately unifying the scene, camera angles, and rendering the animation. The model also used Python scripts to create editable W16 engines and lunar rovers in FreeCAD, and could call external tools like Seedance to create videos. In the official evaluation, Terminal-Bench 4.0 achieved a score of 46.5, outperforming its benchmark models such as Hy4 and DeepSeek V4.1 Flash, but still lagging behind top-tier models like Claude Opus 5.5. The API is currently available on Qianfan. During the limited-time 50% discount, every million tokens costs approximately 1 RMB to input and 4 RMB to output during peak hours, and 0.5 RMB and 2 RMB respectively during off-peak hours. Baidu also promised to open up Preview weights according to the Apache 2.0 license in the future. [Original Link]