GPT-5 vs Gemini 2.5 Pro vs Grok-Code-Fast-1: Agentic Coding Comparison Results

Introduction

I was quite intrigued by grok-code-fast-1, as it scored over 70 on SWE-bench and is cheap and fast. As it's currently free to use, I had it code a simple game and compared it with GPT-5 and Gemini 2.5 Pro.

Comparison Environment

For this comparison, I used the following combinations:

• GPT-5(high) & codex-cli
• Grok-Code-Fast-1 & cline
• Gemini 2.5 Pro & gemini-cli

Results Comparison

Title Screen

First, let's look at the title screens generated by each model.

GPT-5

Grok-Code-Fast-1

Gemini 2.5 Pro

Unfortunately, Gemini 2.5 Pro was unable to display the title screen properly.

Game Screen

Next, let's see the actual game play screens.

GPT-5

Grok-Code-Fast-1

Intermediate Character Design

Let's also compare the character designs generated during development.

GPT-5

Grok-Code-Fast-1

Gemini 2.5 Pro

Coding Performance Summary

Based on this evaluation, here are the impressions regarding the coding performance of each model.

• Code Quality:
  • GPT-5 >> Grok-Code-Fast-1 > Gemini 2.5 Pro
• Implementation Speed:
  • Grok-Code-Fast-1 was the fastest.
  • Gemini 2.5 Pro tended to take longer to implement.
• Design Sense:
  • Grok-Code-Fast-1's generated UI and characters seemed superior in design.

Conclusion

While Grok-Code-Fast-1's coding performance does not match that of GPT-5, considering the cost and response speed, it feels quite attractive. Would it be realistic to replace Grok-Code-Fast-1 for simple tasks or ones that prioritize speed?