llamaperf

DeepSeek V4.1 Flash 552B (16B active)

on M3 Ultra 256GB · Unsloth Studio

Tone: negativeApple hardware
Nov 7, 2026
Throughput
40.0 t/s gen
Quant
IQ3_XXS (GGUF)
System RAM
256 GB

Summary

User reports DeepSeek V4.1 Flash at 40 t/s output on a Mac Studio with 256GB, run through Unsloth at IQ3_XXS. The user finds it slow in practice, saying it takes 5-10 minutes to finish plotting a map that GLM 5.3 Flash finishes in a minute or so. They ask whether others use speculative decoding and what effort level to set.

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.

DeepSeek V4.1 Flash 552B · 16B active at 3-bit by hardware

Bars are medians on one scale, 0 to 73.4 tokens/s. Press ▶ to watch text arrive at a row's speed.

Models most run on the M3 Ultra 256GB

Bars are medians on one scale, 0 to 136.0 tokens/s. Press ▶ to watch text arrive at a row's speed.

Everything reported on the M3 Ultra 256GB →Every DeepSeek V4.1 Flash report →Estimate DeepSeek V4.1 Flash on this card →