AI GUIDEPartner content

Qwen 4 Max creates a game: a practical test of generating, running, and fixing mechanics

A practical test showed that Qwen 4 Max created a playable game prototype with enemies and a final objective in 1 hour and 21 minute. However, the character continued sliding after the key was released, the camera behaved unstably, and the graphics remained simple.

Affiliate link: your price stays the same and the project earns a commission.

Qwen 4 Max created a playable game prototype in this test, but not a finished game. The author checked whether the model could generate a project, transfer it to a server, and enable completion with basic mechanics. The experiment’s main result was that the model was suitable for rapid prototyping; however, the controls, camera, graphics, and individual game scenarios would require further manual refinement.

Task setup and game generation time

At the time of the experiment, Qwen 4 Max was in preview, meaning a pre-release mode. The author selected the highest available reasoning level, X High, and the Alibaba Token Plan, then asked the server to run the created game after generation. The test does not prove that the model can develop a commercial project fully autonomously: a human formulated the task, selected the operating mode, and initiated the launch.

Game generation took 1 hour and 21 minutes. The author then checked not only for the presence of files or an interface, but also the prototype’s actual behavior during gameplay. For game generators, this is a significant test: the code may build without a critical error, while character controls, the camera, and victory conditions only become apparent during interactive execution, when the player can complete the level from beginning to end.

Which mechanics worked after launch

Brief conclusion: the prototype launched and allowed the level to be completed.

According to the author’s observations, the game’s core mechanics worked, including the enemies. The character moved around the level, and completion occurred after reaching a lit area. This result allows the project to be considered a playable prototype rather than a static scene in which objects are merely displayed on screen.

The boundary between “a mechanic is present” and “a mechanic is polished” is especially important here. An enemy may exist and react to the player, but that does not yet mean balanced behavior, reliable collision detection, or a predictable gameplay loop. In the described test, the author confirmed that the basic elements worked, but did not provide a separate assessment of difficulty, replayability, or the quality of the source code.

Character control and sliding problem

The main practical error involved key handling. After a button was released, the character continued running and sliding, making it difficult for the player to stop before the edge of a platform. The author notes that this made it easy to fall.

For a platformer and any level with narrow sections, this kind of error concerns the basic control logic rather than cosmetic shortcomings. A key-release handler should stop acceleration or put the character into a braking state, while the physics model should account for speed, friction, and contact with the surface. In the generated prototype, according to the author’s observation, this behavior was not brought to a comfortable gameplay state.

So what exactly does this defect show? The model was able to assemble the combination of “character plus movement plus level,” but it did not carry out a full check of the user experience of controlling the character. For a rapid prototype, this is an acceptable starting result, but a demo, player testing, or publication would require a separate iteration including checks of stopping, jumping, collisions, and behavior near the level boundaries.

Camera errors and completion conditions

The camera in the game was also unstable. The author reports that it could move downward or move too slowly. As a result, the player lost a convenient view of the level, and the correct direction of movement became less obvious.

Camera problems are especially noticeable in third-person perspective and scenes with vertical changes in elevation. A follow camera should maintain a connection with the character, limit undesirable angles, and respond to position changes without noticeable delay. In the test prototype, these conditions were not always met, so the mechanic’s functionality did not mean comfortable gameplay, especially in areas where the view determines the choice of the next move.

The final condition was simple: completion occurred after the character reached the light. This objective is suitable for a short demonstration and makes it possible to quickly test the gameplay loop from start to finish. It does not show how well the model can design levels with multiple routes, intermediate objectives, failure states, progress saving, and replayability.

Graphics quality and prototype limitations

The author rated the graphics as simple. The visual component served as an environment for testing the mechanics, but did not create the impression of a finished game with a well-developed art style. This matches the nature of the result: Qwen 4 Max assembled an interactive scene faster than a developer could manually prepare a complete set of models, materials, effects, and animations.

When evaluating such a prototype, it is worth distinguishing three levels of results:

  • Launch: the project opens and allows the player to begin the level.
  • Playability: the character, enemies, camera, and objective work together.
  • Quality: the controls are predictable, the camera does not get in the way, the graphics maintain a consistent style, and errors do not break the playthrough.

In the described experiment, the first level was achieved, the second was achieved partially or with noticeable glitches, and the third was not achieved. It is more reasonable to use the result as a foundation for further development rather than as proof that the model is ready to release finished games in a single run.

Unconfirmed similarity to Fable and Gemini generations

The author subjectively noted similarities between the level and mechanics and previously observed generations by Fable and Gemini. This observation concerns the impression created by a specific test, not an established fact about the origin of Qwen 4 Max’s code, design, or training data.

The similarity may be explained by common game-prototype generation patterns. Models often choose similar solutions: a linear route, simple enemies, a contrasting endpoint, and a basic third-person camera. Without comparing the source files, the sequence of prompts, and the results of several independent runs, it is impossible to claim that one model reproduced another’s work.

Is Qwen 4 Max suitable for rapid game prototyping

Based on this author’s test, Qwen 4 Max is suitable for creating a first playable game draft when speed and basic interactivity are important. In 1 hour and 21 minute, the model created a scene, added enemies, and set a completion condition, after which the author launched the project on a server.

The tool does not replace a developer during debugging and polishing. The character controls continued to operate after the key was released, the camera sometimes moved downward or too slowly, and the graphics remained simple. Such errors require not a general request to “fix the game,” but targeted iterations with checks of specific states: releasing each key, stopping at the edge, camera movement at different elevations, collision with an enemy, and reaching the final area.

What should count as success in such a test? If the goal is to obtain an idea that can be opened, played, and then developed further, the result looks useful. If a finished commercial level with reliable controls, expressive graphics, and predictable behavior is required, a single generation cycle is not enough. In the author’s subjective assessment, Qwen 4 Max seemed decent in the experiment, but fell behind DeepSeek and did not look like a flagship-level model.

Test conclusion

Qwen 4 Max did create a playable game prototype with enemies, movement, and a final objective. The experiment confirmed that it is possible to move quickly from a text-based task to an interactive scene, but it also revealed the limitations of automated generation. The model assembled a functional foundation, while a human would have had to fix the character’s sliding, adjust the camera, check collisions, and bring the graphics into a consistent style.

For a developer, this is a useful way to apply the tool: Qwen 4 Max can be viewed as a tool for initial assembly and accelerating experiments. It should be evaluated not by the mere fact that it launches, but by the number of iterations required to achieve reliable controls and a complete playthrough.

Compare models before you start

The service sets its plans, limits and model catalog. If they differ from this article, contact us so we can update it and record a new review date.

Browse models

Affiliate link: your price stays the same and the project earns a commission.

Qwen 4 Max AI game creation neural-network game generation game prototype AI for game developers Qwen 4 Max test game code generation

SEO Mind42 editorial team

We explore SEO and neural networks in practice: test services on our own projects, verify prices and limits against primary sources, and share things you can put to use the same day.

📚 Reference guide to SEO and AI 🔄 Materials are updated 🕐 Updated: 15 November 2026

Related reading

All in this section →