DeepSeek V4.1 Flash Abliterated
on M5 Ultra 256GB · oMLX · 8,192 ctx
Use cases
codingtool-uselong-context
Summary
User reports DeepSeek V4.1 Flash Abliterated at 65.6 t/s decode and 1,857 t/s prefill on an M5 Ultra 256GB, at an 8K prompt with 128 output tokens and zero prefix-cache hits.
Setup is oMLX 0.7.3 with MLX 0.32.4, the oQ2.5e-mtp quant, adaptive MTP, SSD Engram offload, chunked prefill and one active request, with 256K context configured.
Prefill rises to 2,500 t/s at 16K and decode peaks at 79.8 t/s at 200K; initial uncached prefill was 300-700 tok/s and recent code benchmarks reach roughly 1,700-2,500 tok/s. The user is happy with the purchase but notes the gap against specs such as 2x DGX Spark.
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
- 73.4 t/s · one run
- Gemma 4 31B16-bit13.9 t/s · one run
- 74.7 t/s · one run
- 17.0 t/s · one run
- 141.0 t/s · one run
- Qwen3.8 27B4-bit50.0 t/s · one run
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 →