Habr: Comparison of AI Models in Developing Battle City Remake (2026)

The Russian-speaking developer community has conducted a repeat experiment—comparing modern AI models when solving a practical task of developing a video game. This serves as an indicator of real progress in the capabilities of AI systems.

The Original Experiment

Half a year ago, the first article comparing AI models when creating a remake of the classic Battle City game was published on Habr. At that time, the main models of that era were tested.

What Changed in Six Months

• Release of new versions of models (GPT-5.x, Gemini 3+, Claude 3.5+ and others)
• Improvement in code generation capabilities
• Expansion of context windows in most models
• Emergence of specialized models for coding

Practical Utility

Comparison on a real task is more informative than public benchmarks because it shows:

• Developers' ability to work with unfamiliar code
• Capacity for iterative improvement
• Error handling and debugging
• Quality of final working code

What This Means

Repeating the experiment over a short period (6 months) demonstrates an accelerated innovation cycle in AI. For developers, this means the choice of tool is constantly changing, and previous recommendations may already be outdated. It also shows the practical significance of AI for game development and other technical fields.