Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-31 · high
Build b10712 targets Qwen 3.8 Flash Next's sparse attention, and b10714 the same morning tunes AMD's Strix Halo GPU.
Why now: b10712 and its RDNA3 follow-up, b10714, went up the same morning, which is what makes the shift traceable to a single day of releases.
llama.cpp's b10712 release adds a top-k radix sort and top-k QSA fusion path for its Vulkan backend, with tests written against Qwen 3.8 Flash Next. That model's sparse-attention design made large-k sampling the slow part of inference on Vulkan, and the release exists to fix that specific bottleneck.
The same morning, build b10714 went up with a separate fix, RDNA3 mat-vec tuning locked to a static shape of 4 rows above 4 columns, aimed at Strix Halo hardware.
Neither change is a generic backend speedup. Both are written against the shape of one model or one chip. This is a change in how the codebase moves. For years llama.cpp's Vulkan and CUDA work optimized for throughput across whatever model you pointed it at. Now a release exists because one model's attention mechanism has a specific slow path, and another exists because one AMD chip has a specific memory layout.
If that pattern holds, the changelog starts reading less like a backend roadmap and more like a list of which models and chips got attention on a given day. Vulkan users on anything other than Qwen 3.8 Flash Next or Strix Halo hardware are waiting on a dedicated pass of their own, and nothing in these two builds says one's coming.
Each link below shares sources, entities, or timing with this story.
Alibaba released Qwen / Shared entity: Qwen / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (Alibaba released Qwen); both cover Qwen; overlapping topics (architecture, model).
Alibaba released Qwen / Shared entity: Qwen / Shared topic / Earlier coverage
Linked by a graph relationship (Alibaba released Qwen); both cover Qwen; overlapping topics (against, model).
Ollama supports Qwen / Shared entity: Qwen / Same source domain / Earlier coverage
Linked by a graph relationship (Ollama supports Qwen); both cover Qwen; reported by the same outlet (github.com).
Alibaba released Qwen / Shared entity: Qwen / Shared topic / Earlier coverage
Linked by a graph relationship (Alibaba released Qwen); both cover Qwen; overlapping topics (against, architecture).
Linked by a graph relationship (Alibaba released Qwen); both cover Qwen; overlapping topics (architecture, model).
Qwen benchmarked against Claude / Shared entity: Qwen / Earlier coverage / Tension
Linked by a graph relationship (Qwen benchmarked against Claude); both cover Qwen; earlier Qwen coverage from 2026-06-28.
Alibaba released Qwen / Shared entity: Qwen / Earlier coverage / Tension
Linked by a graph relationship (Alibaba released Qwen); both cover Qwen; earlier Qwen coverage from 2026-04-23.
Alibaba released Qwen / Shared entity: Qwen / Same source domain / Shared topic / Earlier coverage
Linked by a graph relationship (Alibaba released Qwen); both cover Qwen; reported by the same outlet (github.com).