More
Best rig by budgetRent before you buySubmit a buildNewsletterFree datasetAdvertise
Submit build

Builds / Apple M5 Ultra 80-core GPU (unified, Mac Studio)

Mac Studio M5 Ultra 256GB: Qwen3.8 Flash-Next 4-bit MTP, 187 tok/s fresh, 94.7 at 32K (TensorFold)

Last updated:

Source: github.com ↗📏 How measured: reported by the original source; RigBase re-checked the numbers at that source
GPU(s)
Apple M5 Ultra 80-core GPU (unified, Mac Studio)
VRAM / unified memory (GB)
256
CPU
Apple M5 Ultra
Motherboard
not stated
RAM
256 GB unified memory
PSU
not stated
Other
macOS 27.0; MLX 0.32.2 / mlx-lm 0.31.3; TensorFold 0.3.4.1 serve --context 262144; thinking off for speed runs
Reported total cost
not stated
Power
not stated

Measured speed (generation tokens/sec)

ModelQuantEnginetok/sPrompt processing (tok/s)Details
Qwen3.8 Flash-NextMLX 4-bit + MTP (Vontra/Qwen3.8-Flash-Next-MLX-4bit-MTP)TensorFold 0.3.4.1 (MLX)187.32,708 @ 2K-token prompt
est. TTFT @8k ≈ 3.0 s (calculated from measured prefill)		⚡ spec-decode (MTP)context: fresh prompt, greedy📏 Decode = wall-time difference between 32- and 512-token completions, on 0.3.4.1 (200.7 on 0.3.4); default MTP drafts, exactness check passes
Qwen3.8 Flash-NextMLX 4-bit + MTP (Vontra/Qwen3.8-Flash-Next-MLX-4bit-MTP)TensorFold 0.3.4.1 (MLX)94.72,324 @ 32K-token prompt
est. TTFT @8k ≈ 3.5 s (calculated from measured prefill)		⚡ spec-decode (MTP)context: after a 32K-token prompt📏 Decode at 32K depth on 0.3.4.1 (sustained figure; 61.0 at 128K, 55.1 at 256K)

Prompt processing = prefill speed (how fast the prompt is read) as reported by the source, with the prompt length it was measured at; blank = the source does not report it. est. TTFT @8k is calculated from the measured prefill speed (8,192 ÷ prefill tok/s), shown only when prefill was measured on a prompt of 512+ tokens. It is not a measurement and ignores load time and caching. measured TTFT is the source's own time-to-first-token. Why prefill matters for coding assistants.

Independent run posted on TensorFold issue #22 (27 Sep 2026, on behalf of @sethforprivacy). Peak RAM 135 GB. Same box: oMLX 0.7.0rc1 (oQ6e) 147 tok/s fresh, mlx-serve 160 tok/s fresh. Concurrent requests queue rather than batch (aggregate flat at ~180-195 tok/s from 1 to 8 streams).

Source: https://github.com/ashhart/TensorFold/issues/22

Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.

Embed this result

Copy into a README or blog. Links back to this build. See badge docs (CC BY 4.0 attribution). The SVG does not invent speeds — numbers stay on the build page.

Markdown

[![Measured on RigBase](https://getrigbase.com/badge/mac-studio-m5-ultra-256gb-qwen38-flash-next-tensorfold.svg)](https://getrigbase.com/builds/mac-studio-m5-ultra-256gb-qwen38-flash-next-tensorfold/)

HTML

<a href="https://getrigbase.com/builds/mac-studio-m5-ultra-256gb-qwen38-flash-next-tensorfold/" rel="noopener"><img src="https://getrigbase.com/badge/mac-studio-m5-ultra-256gb-qwen38-flash-next-tensorfold.svg" alt="Measured on RigBase" height="20"/></a>

Preview: Measured on RigBase · How to use

Rent before you buy: try this model on a cloud GPU for a few dollars first: Runpod (referral link; sign up with Google to get the new-user bonus)

Quick answers: Can Apple M5 Ultra run Qwen3.8 Flash-Next?