The ggml‑org/llama.cpp project released a new version on 2026-11-13. The update deletes several deprecated APIs, including llama_set_warmup and the defrag_thold option. It also moves the speculative NextN API from the auxiliary llama‑ext.h file into the primary llama.h header and renames related functions. Documentation for the llama_model_dflash_selector_top_k function was added.
Why it matters
Developers can now use the NextN speculative generation directly from the main API, simplifying code and reducing reliance on outdated functions.
