The open‑source llama.cpp library now stores mixture‑of‑experts (MoE) data in a GPU cache kept in host memory.
NoteBy OossaPublished by Oossa: Last updated: 1 min read
The ggml‑org team released llama.cpp version b11480 on 2026‑10‑08. The update adds a GPU cache for MoE experts that lives in host memory, helping large models run faster on GPUs. The change is marked as assisted‑by Claude and uses a new llama_moe_cache_ptr API. Pre‑built binaries for macOS, iOS, and several Ubuntu configurations are available for download.
Developers can now run MoE models more efficiently on consumer GPUs, lowering the hardware cost for AI projects.
Models
The company’s new three‑agent Claude‑based reviewer found 73.43% of core‑claim errors, far above the previous best of 14.81%.
Models
Mistral's new trillion‑parameter model scores 38 on an independent index, but costs more per task than competing open models.
Models
OpenAI reported two influence campaigns that used AI‑generated content to plant false stories in real media and blocked the involved accounts.
Oossa · Newsletter
The week in AI, explained
Every Monday: the stories worth knowing, in plain language. Free, no spam.
