Strata 0.2.8 was released on October 29, 2026, introducing support for faster Intel Arc B65 cards, a four-card Radeon setup that significantly increases throughput, and a fix for a memory leak affecting Mac users. This open-source engine can run Alibaba's 125-billion-parameter Qwen3.8-Flash-Next model on consumer-grade PCs with as little as 12 GB of VRAM, leveraging GPU-RAM-SSD splitting and speculative decoding techniques. The project, which has quickly gained around 36,000 GitHub stars in five weeks, is primarily developed by Niko Veit and emphasizes transparency in performance metrics and limitations. While the updates provide notable improvements in specific configurations, users should be aware of trade-offs in quality due to aggressive quantization and the varying performance based on hardware setups

Strata 0.2.8 Released: Four-Card Radeon Support, Bigger Local AI Performance

Strata 0.2.8 shipped on October 29, 2026, adding faster paired Intel Arc B65 cards, a four-card Radeon setup that triples throughput, and a fixed memory leak that was crippling DeepSeek on Macs. The open-source engine runs Alibaba's 125-billion-parameter Qwen3.8-Flash-Next on a PC with as little as 12 GB of VRAM, using clever GPU-RAM-SSD splitting and speculative decoding. Not all gains are universal though, since the author stresses every speedup is tied to a specific configuration and aggressive quantization quietly trades away quality. The project grew to roughly 36,000 GitHub stars in five weeks thanks to its free, offline, privacy-first approach, even though it is largely a solo effort by Slovenia's Niko Veit.

Strata 0.2.8 Released: Four-Card Radeon Support, Bigger Local AI Performance @ Linux Compatible