Research

Studies we run while building our products, published with their method, their results and their limits.

• Six models author a logic puzzle: which ones get through the game's own gates?
  September 2026 · GRIDIGMA · Language models from Anthropic, OpenAI and xAI author the same ten puzzles through the game's production loop; passes, hardness, time and a blind read.
• Nine models, one puzzle game: who writes the clearest English and the most natural Turkish?
  September 2026 · GRIDIGMA · Language models from Anthropic, OpenAI and xAI rewrite puzzle text, translate it into 21 languages, then rate each other blind.