Kimi K2.6, an open-weights Chinese AI model from Moonshot AI, stunned the tech world by defeating Claude Opus 4.7, GPT-5.5, and Gemini Pro 3.1 in a real-time programming contest. The Word Gem Puzzle challenge required models to solve sliding-tile letter grids under strict scoring rules, with Kimi K2.6 securing 22 match points through aggressive tile manipulation. Unlike static-scanning opponents, Kimi prioritized dynamic word creation, unlocking high-value combinations through relentless grid traversal.
Its strategy faltered on smaller grids but dominated 30×30 boards where seed words disintegrated, forcing competitors to adapt.
Source: Hacker News · Summarized by HeadlinesBriefing