Quick initial impressions: - at 70 tok/s slower than GLM 5.1 - seems to spend more time reasoning - better results with my Pac-Man test The one-shot result is almost functional; apart from the ghosts getting stuck immediately after leaving the ghosts house, I did not notice any other obvious bug. Everything else seems to function much better than any of the other models I tried, and the game is more complete. This ranks it in first place . Second place is Qwen 3.6 27b. You can test th...