Strata v0.1.38 runs Qwen3.8-Flash-Next — a 125B MoE model — on a gaming GPU at 94 tokens/sec. Here’s how MoE makes it possible, and what the benchmarks don’t tell you.
The post Strata Runs a 125B AI Model on Your Gaming PC — No Server Needed appeared first on byteiota .
source & further reading
byteiota.com — original article
Janus: Run Local LLMs From a Single Go Binary — No Python, No Docker
OpenAI Dots: The Always-On Agent That Never Sleeps
D-Engine: 14x Fewer Tokens, Zero Broken Commits