The author describes hands-on experience with Hy4 GA across image generation and two types of 3D scenarios: it produced a relatively complete industrial plant, but slowly and with mediocre detail and aesthetics; its sandbox game had a distinctive visual style, but demonstrated less complete mechanics than DeepSeek V4.1 Flash. The article provides no reproducible inputs or run configuration, so its conclusions reflect only the author's observations of these examples.
Tasks this can help assess: The author's experience with a pelican image prompt, a virtual industrial plant, and 3D block-based sandbox generation.
Tasks this should not be generalized to: Comparing overall model capability, success rate, speed, or cost; assessing consistent performance across different prompts and tool environments.
Model version covered: The author says Hy4 GA; no API model ID or snapshot is provided. The comparison model is DeepSeek V4.1 Flash.
Test environment or client: Not specified. The industrial plant example is said to have been generated from a prompt; whether it was run through WorkBuddy, CodeBuddy, or an API is not stated.
Reasoning tier and parameters: Not specified.
The article describes no standardized evaluation plan or unified scoring. The author recounts three hands-on trials:
Used the familiar “pelican test” to observe Hy4's image-generation result and visually compare it with Hy4 preview and DeepSeek V4.1.
Generated a virtual Tesla factory scene from a prompt; the author says the model also handled the camera movement in the screen recording.
Compared interactive 3D block-based sandbox games generated by Hy4 and DeepSeek V4.1 Flash, looking at visual style and game mechanics.
Pelican image prompt: The author considers Hy4 GA better than the preview, but still visually behind DeepSeek V4.1. The article lists no prompt, image input, generation parameters, or evaluation criteria.
Virtual factory: The author describes an output with a 3D plant, robotic arms, AGV carts, and interactions such as selecting equipment, checking its status, switching camera views, and visual monitoring. The author considers it relatively complete overall, but says generation was slow, some modules were mediocre, and the aesthetics need improvement. No generation time or feature acceptance record is provided.
Block-based sandbox: The author says both models generated interactive games. In the author's description, DeepSeek had more complete terrain and mechanics for destruction, placement, and material switching; Hy4 stood out for its cartoon style, grass, flowers, trees, and sunset sky. The article does not disclose the game-task prompts, code, or evidence that both ran in the same environment.
The article provides no independently measured quantitative test data. The results above are the author's textual descriptions and subjective assessments of the examples; they should not be treated as blind-test or repeated-experiment results. The Artificial Analysis rankings, model prices, and Tencent use-case information in the article are background material and are not part of this set of hands-on results.
This article adds image and 3D generation experiences to the existing coverage of coding, office, and planning tasks, and offers the author's observations on visual style, generation completeness, and speed. The evidence is personal experience: the article specifies no test environment, prompts, model invocation method, generation times, scoring rubric, or number of repetitions, and provides no run logs that can be independently checked. The sandbox comparison is especially dependent on task inputs and the generation environment, so it cannot establish the models' general game-development capabilities.
This note summarizes publicly available content from the original article; it is not an independent retest. Reproduction would require the complete inputs for all three tasks, model and client configurations, generated artifacts, and run processes, followed by repeated tests under the same conditions. The original article provides none of these materials.
Hy4