llamaperf

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

on 2× NVIDIA T4 16GB · Strata · 196,608 ctx

Tone: positiveNVIDIA hardware
Nov 22, 2026
Throughput
38.7 t/s gen · 1160.0 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
UD-Q4_K_XL (GGUF)

Use cases

agentictool-uselong-context

Summary

User reports Qwen3.8-Flash-Next at 38.7 t/s decode and ~1,160 tok/s prefill on a 189,104-token prompt with a 196,608-token live context, on 2x Nvidia Tesla T4 16GB. Setup is Strata with UD-Q4_K_XL GGUF, expert cache in VRAM and spill to host RAM, --kv-resident 32768. The same user runs Qwen3.6-35B-A3B UD-Q4_K_M in llama.cpp at 55 t/s decode (up to 70 depending on context size and reasoning effort) with ~1k t/s prefill at 237,568 context, entirely in VRAM. Peak GPU power during the context test was 149W, with the server averaging ~320W.

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.

Everything reported on the NVIDIA T4 16GB →Every Qwen3.8 report →Estimate Qwen3.8 on this card →