Kimi K2.6 Beats Claude, GPT-5.5, and Gemini in Coding Challenge

Moonshot AI's open-weight model Kimi K2.6 took first place in the Word Gem Puzzle category of the 'AI Coding Contest.' In this competition, Chinese models swept the top rankings, surpassing models from major Big Tech companies.

Contest Results and Rankings:

• 1st: Kimi K2.6 (Moonshot AI)
• 2nd: MiMo V2-Pro (Xiaomi)
• 3rd: GPT-5.5 (OpenAI)
• 4th: GLM 5.1 (Zhipu AI)
• 5th: Claude Opus 4.7 (Anthropic)

Word Gem Puzzle is a sliding-tile character puzzle where finding longer English words within the grid earns higher scores. As the grid size increases (up to 30x30), existing words get randomly scrambled, making it important not just to find existing words but to actively move tiles to create new words.

Differences in Model Strategy:

• Kimi K2.6: Recorded the highest score using a Greedy Sliding strategy that continuously pushes tiles to open up new words.
• MiMo V2-Pro & Claude: Relied on static scanning, finding words in the initial grid without directly moving tiles. This approach was effective on small grids but revealed its limits on the 30x30 grid, where it could no longer secure additional words.
• GPT-5.5: Used a conservative sliding approach that avoids infinite loops, showing strength on larger grids.

While this result reflects performance on a specific puzzle, it suggests that the open-weight model Kimi K2.6 is delivering performance on par with major closed models, further narrowing the competitive gap between models.