Honestly? I'm betting that within a couple months, Gemini's gonna dominate the coding model space. The pretrained foundation they've built is seriously impressive—there's something unusual about how it performs. Hard to put my finger on it, but the architecture feels... different. Not your typical transformer behavior. Could be the training data mix, could be some novel attention mechanism. Whatever it is, the early signals are weird in a good way. If they nail the fine-tuning phase, we might see a real shift in which AI devs actually reach for when building.