← sources

[Benchmark] Strata/Basalt runs Qwen3.8 Flash Next IQ3_S with 16GB VRAM and 55GB RAM with up to 250 tok/s generation and 5,600 tok/s prompt processing (5080, Concurrency 1)

source: reddit (reddit:LocalLLaMA)published 2026-10-25 03:59ingested 2026-10-26 02:40

retained text

Qwen3.8-Flash-Next IQ3_S on Basalt | RTX 5080 16 GB, Core Ultra 9 285K, 62 GB DDR5-6800 | 262K context, int8 KV

| Workload (prompts 5K+) | PP avg (t/s) | PP peak (t/s) | TG avg (t/s) | TG peak (t/s) | VRAM | RAM (engine / left free) |
|:--|--:|--:|--:|--:|--:|--:|
| Agent coding session (opencode, 103 requests, up to 130K ctx) | 5,011 | 5,118 | 153.0 | 262 | 15.7 GB | 54.1 GB / 5.2 GB |
| Context sweep (fresh prompts 15K-207K) | 5,475 | 5,662 | 154.9 | 186.6 | 15.7 GB | 54.1 GB / 5.1 GB |

* Agent PP: the two big uncached reads (10.7K and 8K new tokens). 99.5% of prompt tokens came from the prompt cache; median time to first token 0.2 s.
* Agent TG peak is the best single second; the average is all generated tokens over all decode time.
* Sweep TG is per-request averages (384 tokens each, temp 0.7), 2 runs per depth.

document fetched 2026-10-26 00:38 · canonical https://reddit.com/r/LocalLLaMA/comments/1xesda1/benchmark_stratabasalt_runs_qwen38_flash_next_iq3