The llama.cpp project released version b11488 on 2026-10-08. It adds the LiquidAI/d1-omni-600M decision model, renames the decision type to lfm2-d1-omni, and updates the server to handle audio via images and an alias for files. The changes also keep a specific neural‑network activation (conformer GLU sigmoid) on CUDA. These updates are documented in the GitHub release notes.
Why it matters
Developers using llama.cpp can now run the new d1‑omni model for decision tasks without extra code changes.
