248k-context

Gemma 4 26B runs on an RTX 4060 with 248K token context at 20 tokens per second using llama.cpp and Q4_K_XL quantization, enabling local processing of entire codebases on consumer hardware.