Gemma 4 12B runs locally without compromise

Google DeepMind's Gemma 4 12B runs locally on a laptop without feeling like a compromise. The 12-billion-parameter model handles multi-step reasoning and a 256K context window. Running it via LM Studio on a MacBook Pro delivered zero-latency, private, offline performance that felt like an upgrade over cloud-dependent alternatives.