llamaperf

DeepSeek V4.1 abliterated

on M5 Ultra 256GB · oMLX · 1,025 ctx

Apple hardware
Nov 5, 2026
Throughput
55.3 t/s gen · 640.8 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.
System RAM
256 GB

Use cases

coding

Summary

User benchmarks a DeepSeek V4.1 abliterated model on a single M5 Ultra 256GB Mac using oMLX with the Auto engine and a Python code prompt. At a 1025-token prompt the run reaches 640.8 tok/s prompt processing and 55.3 tok/s generation; longer prompts raise generation to 62.1-81.3 tok/s across the sweep up to 200001 tokens. Continuous batching at pp1024/tg128 gives 118.2 tok/s aggregate at 2x and 220.0 tok/s at 4x, with peak MLX memory around 210.8 GB.

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 report →Estimate DeepSeek V4.1 on this card →