The Oct 10, 2026 release fixes a bug where a new request could corrupt an ongoing generation by rewriting its prompt cache.
NoteBy OossaPublished by Oossa: 1 min read
The llama.cpp project released version b11552 on 2026‑10‑10. The change stops a request from updating the prompt cache of a slot that is already busy. Previously, the busy slot would be overwritten, causing the generation to continue with the wrong context. Now the busy slot is returned unchanged and the new request waits until the slot is free.
Developers using llama.cpp will see fewer generation errors when multiple requests share the same GPU memory.
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.
Models
Anthropic's AI workbench stitched together decades of space‑telescope data and filled gaps, producing a complete UV map of the sky.
Oossa · Newsletter
The week in AI, explained
Every Monday: the stories worth knowing, in plain language. Free, no spam.
