The open‑source LLaMA inference library releases version b11510 with a new CUDA kernel that handles more than 65,535 rows, plus binaries for macOS, iOS and Linux.
NoteBy OossaPublished by Oossa: 1 min read
The ggml‑org team posted llama.cpp version b11510 on Oct 8 2026. It adds a looped PAD kernel for CUDA so the library can process tensors with over 65,000 rows or slices. The release also bundles pre‑built binaries for Apple Silicon, Intel macOS, iOS, and several Ubuntu builds (CPU, Vulkan and CUDA 12.8). The code and attestations are linked from the GitHub page.
Developers can now run larger language models on NVIDIA GPUs without hitting the previous row limit.
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.
