The open‑source llama.cpp library now uses AVX512‑FP16 to compute half‑precision dot products in full precision, improving CPU performance.
NoteOossa1 min read
The ggml‑org team released llama.cpp version b11262 on Sep 29, 2026. The update adds AVX512‑FP16 support, which does half‑precision (f16) dot products but accumulates them in single‑precision (f32) for better accuracy. It is a low‑level change that speeds up inference on CPUs that have the AVX512‑FP16 instruction set. The release also bundles new binaries for macOS, iOS, and several Linux builds.
It lets developers run LLMs faster on modern CPUs without losing numerical quality.
Models
OpenAI’s new GPT‑6.1 Sol keeps the $2‑$10 per‑million‑token rate while claiming stronger performance, undercutting Anthropic’s Claude and DeepSeek models.
Models
The new v4 and v4 Turbo voices cost $0.08 and $0.04 per 1,000 characters, with a promotional cut to $0.022 and $0.011 per 1,000 until Oct 12.
Models
Oracle’s new Fusion Claw runtime lets its Fusion apps handle multi‑step jobs like staffing and accounting, using AI for reasoning while keeping calculations outside the model.
Oossa · Newsletter
The week in AI, explained
Every Monday: the stories worth knowing, in plain language. Free, no spam.
| # | Source | Outlet | Date | Key takeaway |
|---|---|---|---|---|
| 1 | ggml-org/llama.cpp b11262 ↗ | llama.cpp | Sep 29, 2026 | <details open> ggml : accumulate f16 dot products in f32 on AVX512-FP16 (#29545) Supersedes #29530 Signed-off-by: Adrien Gallouët <angt@hugg |
1 sources
