Kimi K2.6 just beat Claude, GPT-5.5, and Gemini in a coding challenge (thinkpol.ca)

In a surprising turn of events, Kimi K2.6, an open-weights model developed by the Chinese startup Moonshot AI, triumphed over leading models like Claude, GPT-5.5, and Gemini in a recent coding challenge. The contest, which analyzed performance in a complex Word Gem Puzzle, involved ten AI models competing in real-time programming tasks with scoring based on the length of words formed with letter tiles. Kimi K2.6 achieved an impressive 22 match points with a strategy focused on aggressive sliding and word claiming, illustrating the potential of open-access AI models in a competitive landscape dominated by Western counterparts. The outcome signifies a notable shift in the AI/ML community, as it suggests that open-weights models can now compete closely with established models from major labs. Kimi’s success—paired with MiMo V2-Pro’s second place—highlights a growing trend wherein freely available models are effectively challenging proprietary systems. The challenge also revealed significant discrepancies in strategies: Kimi’s greedy approach to word scoring performed well on larger grids, while several models, including Claude and GPT-5.5, struggled with a static scanning strategy as the grid size increased. This event underscores the evolving dynamics in AI model capabilities and the importance of adaptability in AI training methodologies.