llamaperf

DeepSeek V4.1 Flash Abliterated

on M5 Ultra 256GB · oMLX · 262,144 ctx

Tone: positiveApple hardware
Oct 31, 2026
Throughput
75.0 t/s gen · 1600.0 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
oQ2.5e (MLX)
MTP (Multi-Token Prediction)
on
System RAM
256 GB

Summary

User reports DeepSeek-V4.1-Flash-1017-Abliterated at 1600 tok/s prefill and 75 tok/s decode on an M5 Ultra 256GB. Setup is oMLX with oQ2.5e MLX weights (232GB), MTP and CED enabled, 256K context window. Out of box prefill was 300 tok/s before settings tweaks; user plans to try 2x Thunderbolt 5 when a second M5 Ultra arrives.

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.

Models most run on the M5 Ultra 256GB

Bars are medians on one scale, 0 to 141.0 tokens/s. Press ▶ to watch text arrive at a row's speed.

Everything reported on the M5 Ultra 256GB →Every DeepSeek V4.1 Flash report →Estimate DeepSeek V4.1 Flash on this card →