#google

Downloaded the 27B variant last night. Quantized to Q4 it fits in 16GB RAM and generates at a very usable speed. Not Claude-level reasoning, but for a local model it's a big step. Google is winning the 'open weights you can actually run' game right now.