Developer Benchmarks Qwen3.8 Models on Consumer GPUs

The developer of the capy-flash engine benchmarked Qwen3.8-27B and Flash-Next (125B MoE) on consumer hardware, reaching 540 tok/s on a single RTX 5090 and 379 tok/s on dual 5070 Ti, noting RAM costs can exceed GPU costs.

2026-10-19 ~ 2026-10-19 · 2 related posts

• Verified engine map for Qwen3.8-27B on consumer GPUs: 379 tok/s on two 5070 Tis, RAM costs more than the card — FeydRowan · 2026-10-19
• Verified local LLM engine benchmarks: Qwen3.8-27B up to 540 tok/s on a single RTX 5090 — FeydRowan · 2026-10-19