AI news · October 5, 2026
Strata brings Qwen 3.8 Flash Next to consumer GPUs

Strata is an open-source runtime for Qwen 3.8 Flash Next that targets ordinary gaming PCs. Its GitHub README lists Windows and Linux support, a minimum of 12GB of GPU memory, optional image input and measurements ranging from 44 to 94 tokens per second on example RTX 5070 and RX 9070 systems. Hacker News placed the project on its October 4 front page.
That combination lowers the cost of trying a large model locally and gives builders a direct way to test latency, privacy and hardware limits. It also shifts attention from model claims to the serving stack and the machine that actually runs the workload.
Start with the repository's supported hardware and reproduce its benchmark on your own prompts. Compare memory use, sustained speed and output quality before deciding whether a local deployment can replace an API for your workload.
Source: GitHub ↗ — Made With Models writes the brief; the reporting is theirs.