Strata’s latest v0.2.7 release entry now includes 0.2.7.1 notes for DGX Spark and Apple silicon DeepSeek support. GitHub dates the entry to October 27 and its latest edit to October 28. We checked it October 28; the patch label and the tag name are intentionally different.
The point for existing 12 GB workstations
The maintainer explicitly says this patch brings no new engine or speed change to Windows, x86 Linux, or Mac users running Qwen only. Control advances to 0.1.53, while existing 0.2.7 engine packages remain at 0.1.52. An updater improvement is not evidence that a previously measured generation benchmark has increased. Our published workstation tests remain separate.
Spark is a specific target, not all ARM computers
The DGX Spark guide accompanies support for Qwen3.8-Flash-Next and DeepSeek-V4.1-Flash on that device. The release reports testing one Spark and requires recent Ubuntu/glibc and NVIDIA driver versions. Its Linux ARM path is terminal-based, without a desktop app. Do not translate that support statement into a promise for an ordinary ARM laptop or a CPU-only board.
DeepSeek on the Mac still needs a lot of storage
The Mac guide describes a separate Metal engine for Apple silicon and macOS 15 or newer. The 2.00 and 2.25 bpw sets require 257 GB and 274 GB of disk respectively. The project recommends at least 64 GB of memory but reports measurements from one M5 Max with 128 GB. The supported path is text-only; the weights do not fit wholly in memory and expert data is streamed from fast storage.
For a short code prompt, the guide reports 15.2–16.0 generated tokens/s for the 2.00 bpw set. That measurement had a separate, loaded-but-idle Qwen server sharing the same machine. It is the maintainer’s result under those conditions, not an AiBrainWorX test or a forecast for smaller Macs.
A useful release without a universal speed claim
The practical gain is a wider set of documented hardware choices. Before updating, check which component and model family actually changed, the available disk space, and the architecture supported by that engine. We have not installed this patch, downloaded DeepSeek weights or altered the launcher during this news check.
Original AiBrainWorX reporting and analysis. Primary sources checked October 28, 2026. Publisher demonstrations and benchmarks are not our independent tests.
