Strata 0.2.11 was released on November 5, 2026, introducing several enhancements, including seven new compressed model files that enable running Alibaba's Qwen3.8-Flash-Next on systems with as little as 12 GB of VRAM. This update significantly improves prompt processing speed by up to 70% on NVIDIA cards and includes a fix for a privacy issue that previously caused replies to be sent to the wrong user. Additionally, conversations are now saved as files instead of being stored in browser memory, enhancing data management and security. Overall, this version marks a substantial advancement for the Strata project, although it still has limitations in low-bit quality and model performance on certain hardware configurations
Strata 0.2.11 Adds Compressed Qwen3.8-Flash-Next Files, Faster Prompting, and a Privacy Fix
Strata 0.2.11 shipped on 5 November 2026, adding seven new Unsloth model files compressed from roughly 1-bit to 4-bit that let you run Alibaba's 125-billion-parameter Qwen3.8-Flash-Next on hardware with as little as 12 GB of VRAM. The release also speeds up prompt processing by up to 70% on NVIDIA cards and 32.7% on four Radeon AI PRO R9700 cards, while fixing multi-card starts that previously failed outright. A standout change is a privacy fix for issue #2892 that stops replies from being routed to the wrong user, along with conversations now saved as files instead of browser storage. It's the most feature-rich release from Niko Veit's fast-rising project, though the numbers are self-reported and low-bit quality still has limits.
