Qwen3.8 Flash Next at IQ3_S hits 275 tok/s on a single RTX 5080 via Basalt fork
yehyakar · reddit · 2026-10-29
Reddit user yehyakar set a new record on Basalt (a Strata fork): running Qwen3.8 Flash Next at IQ3S quantization on a single RTX 5080 (16GB, PCIe 5.0 x16), 64GB DDR5 6400 RAM and a Core Ultra 9 285K, they measured 5,500 tok/s prompt processing and up to 275 tok/s peak generation during a live opencode coding session—code/tool calls peaking at 275 tok/s and reasoning traces around 190 tok/s. The author says IQ3S is "super usable"; some of their optimization PRs (xyehya) are still pending upstream, which already includes solid gains and memory optimizations. Code is on GitHub.
More from Infra
- PGlite crosses 30 million weekly downloads as in-browser Postgres momentum builds — matei_zaharia · 2026-10-29
- Prime Inference scales from zero to 120B tokens per day on OpenRouter in one month — xeophon · 2026-10-29
- SJTU's TAO Cuts LLM Agent Serving Latency Up to 26.8% With Tool-Aware Joint Scheduling — AIOR-Research · 2026-10-29
- Nscale's unproven CEO chases $30 billion neocloud IPO amid AI turmoil — bloomberg · 2026-10-29
- Google Cloud CEO: Apple Scaling Foundation Model Pretraining on Latest TPUs — shayan_ · 2026-10-29
- FlashAttention vs standard attention in one chart: the bottleneck is data movement, not compute — cneuralnetwork · 2026-10-29