Claude vs. Gemini vs. ChatGPT vs. Grok (boreal.social)

In a recent coding challenge, AI models Claude, Gemini, ChatGPT, and Grok were tested in a real-time competition to build a Python client that interacts with a TCP server to find words in a grid. Claude emerged as the clear winner, scoring a total of 854 points across three rounds, while the other models struggled significantly. Notably, ChatGPT recorded a staggering cumulative score of -74,383 by submitting a multitude of valid but short words that cost it points. Grok and Gemini, facing their own challenges, found themselves unable to retrieve any points, with Gemini finishing with zero due to a slow submission process. This competition highlights important lessons for the AI/ML community, particularly regarding the need for precise interpretation of specifications and optimization in coding architecture. All three losing bots failed to apply the crucial scoring formula, treating the grid’s minimum word length of three letters as their submission threshold rather than focusing on profitability. Claude's design cleverly implemented a three-thread system that optimized word submissions, whereas the others used a synchronous model that stymied their efficiency. The results emphasize the importance of comprehending requirements fully and structuring code to maximize performance, particularly in competitive environments.