llama.cpp lands MiMo-V2.5 text support

llama.cpp keeps turning “supported in llama.cpp” into the practical launch pad for new open-weight models. MiMo-V2.5 is a useful stress test for the runtime, but the current patch is mostly about making the text path usable before the rest of the stack catches up.

• –MiMo-V2.5 is enormous on paper, so the real value here is quantized/local experimentation, not casual laptop inference
• –Text-only support matters because it lets downstream GGUF builds and tooling move before full multimodal support lands
• –The model’s 1M context and sparse-MoE design make it interesting for long-horizon workflows, but also harder to serve cleanly
• –This is the kind of compatibility work that keeps llama.cpp relevant as the default runtime for new open-weight architectures