llamaperf

Qwen3.8 125B (6B active) Flash-Next

on AMD RX 7900 XTX · Strata · 262,144 ctx

Tone: positiveAMD hardware
Nov 2, 2026
Throughput
86.0 t/s gen · 3100.0 tokens/s prompt processing (prefill)Prompt processing is how fast the input is read, not how long you wait for the first word. How to measure time to first token.
Quant
IQ3_S (GGUF)
KV cache
oct3
VRAM reported
24 GB

Use cases

agenticlong-context

Summary

User reports Qwen3.8-Flash-Next at 86 tok/s median decode and ~3.1k tok/s median prefill on a single RX 7900 XTX 24GB. Setup is Strata with IQ3_S weights and an oct3 3-bit KV cache, 4-token MTP speculative decoding, full 262K context resident on the card. Figures come from 8.7k real requests in the engine log, not a synthetic benchmark; decode reaches 107 tok/s at p90 and 154 peak, prefill up to 3.6k, and a 250K-token prompt is read in ~80 s. The user says oct3 shows no noticeable quality drop versus k8v4 in agentic use.

Compare this run

Medians of plain runs: one device, one request, no speculative decoding, the whole model in the device's memory. The report above is one setup, so read it beside them and not as one of them.

Qwen3.8 125B · 6B active at 4-bit by hardware

Bars are medians on one scale, 0 to 232.0 tokens/s. Each dot is one run and opens its report. Press ▶ to watch text arrive at a row's speed.

Models most run on the AMD RX 7900 XTX

Bars are medians on one scale, 0 to 142.8 tokens/s. Each dot is one run and opens its report. Press ▶ to watch text arrive at a row's speed.

Everything reported on the AMD RX 7900 XTX →Every Qwen3.8 report →Estimate Qwen3.8 on this card →