Models

This page was written and translated automatically by AI and checked by software against its sources, with no human editing. Specifications come from the sources listed; memory requirements and speeds are estimates calculated with published formulas. How we use AI

Parameters and file sizes come from the model repositories on Hugging Face; the architecture is read from the GGUF file header that llama.cpp itself loads.

Values updated Oct 19, 2026

Parameters count the tensors llama.cpp loads (multi-token-prediction layers are not counted). Active parameters per token count the non-expert tensors and the experts in use but not the input embedding lookup; a tied input/output embedding is counted once, as the output layer. Model cards often count these differently, so their figures can be higher. Methodology