llama.cpp hits v1.0 with FP8 KV cache, host-memory MoE cache, and 35 removed deprecated APIs
mark_k · x · 2026-11-14
llama.cpp has reached v1.0.0, with the underlying ggml library hitting 1.0 at the same time — a major release for local AI inference.
Key additions:
- OCP FP8 KV cache and a host-memory MoE expert cache with a configurable CPU share, plus loading of external PLE shards
- Three new text architectures (Kimi-K3.1-Pro, Ripple, xing4.0), drama-3 TTS, and MageVL-Cowork / Kimi-K3.1 vision support
- Hardened llama-server (SSRF, CORS, DNS rebinding, MCP tool cancellation) and a new OpenAI-compatible /v1/audio/speech endpoint
- A redesigned web UI with chat groups and a markdown editor
Upgrade caveat: the release removes 35 deprecated API functions along with options like --defrag-thold, so developers should audit their calls before upgrading.
Related event: llama.cpp hits v1.0.0, giving local AI a stable foundation(15 posts)→
More from Infra
- NIDO: an open-source Android AI assistant that runs fully offline, down to the model and memory — NIDO_007 · 2026-11-14
- Cloud Reports Measure the Wrong Thing: Hybrid Cloud Is Often Legacy, Not Strategy — DavidLinthicum · 2026-11-14
- Will Depue claps back at Teortaxes over claim that 2-3K Blackwell runs aren't "data-center scale" — teortaxesTex · 2026-11-14
- Local LLM stack cheat sheet: seven open models mapped to chat, coding, TTS and more — alex_verem · 2026-11-14
- Qwen3.8-Flash NVFP4 wows on V100s: insanely fast prefill with Strata — Scobleizer · 2026-11-14
- How a $1M 'fake chip' saved Nvidia in 1997 — the RIVA 128 bet that made Jensen Huang — bigaiguy · 2026-11-14