Builds / Apple M5 Ultra 80-core GPU (unified, Mac Studio)
Mac Studio M5 Ultra 96GB (oMLX oQ8e + Lightning MTP): Inkling 2 Mini at 83.4 tok/s at 4k context, 56.2 tok/s at 32k
Last updated:
- GPU(s)
- Apple M5 Ultra 80-core GPU (unified, Mac Studio)
- VRAM / unified memory (GB)
- 96
- CPU
- Apple M5 Ultra (80-core GPU)
- Motherboard
- not stated
- RAM
- 96 GB unified memory
- PSU
- not stated
- Other
- oMLX v0.8.1, macOS 27.2.1; Inkling-2-Mini-oQ8e; Lightning MTP on; Affine8 8-bit KV (turboquant); benchmark context Code (Mixed); temperature 1
- Reported total cost
- not stated
- Power
- not stated
Measured speed (generation tokens/sec)
| Model | Quant | Engine | tok/s | Prompt processing (tok/s) | Details |
|---|---|---|---|---|---|
| Inkling 2 Mini | oMLX oQ8e, Affine8 8-bit KV | oMLX v0.8.1, Lightning MTP | 76.6 | 2,547 @ 1k-token prompt est. TTFT @8k ≈ 3.2 s (calculated from measured prefill) | ⚡ spec-decode (MTP)context: 1k tokens📏 Public oMLX session on an M5 Ultra (80-core) with 96 GB unified memory. Peak mem 20.0 GB at this context. Prefill (PP) and generation (TG) as published on the session page. Retrieved 21 Nov 2026. |
| Inkling 2 Mini | oMLX oQ8e, Affine8 8-bit KV | oMLX v0.8.1, Lightning MTP | 83.4 | 3,225 @ 4k-token prompt est. TTFT @8k ≈ 2.5 s (calculated from measured prefill) measured TTFT: 1.3 s @ 4,096-token prompt | ⚡ spec-decode (MTP)context: 4k tokens📏 Same oMLX session as the 1k row. Peak mem 22.2 GB. TTFT 1,270 ms as published on the 4k result page. Retrieved 21 Nov 2026. |
| Inkling 2 Mini | oMLX oQ8e, Affine8 8-bit KV | oMLX v0.8.1, Lightning MTP | 56.2 | 2,968 @ 32k-token prompt est. TTFT @8k ≈ 2.8 s (calculated from measured prefill) measured TTFT: 11.0 s @ 32,768-token prompt | ⚡ spec-decode (MTP)context: 32k tokens📏 Same oMLX session as the 1k row. Peak mem 28.3 GB. TTFT 11,040 ms as published on the 32k result page. Retrieved 21 Nov 2026. |
| Inkling 2 Mini | oMLX oQ8e, Affine8 8-bit KV | oMLX v0.8.1, Lightning MTP | 74.7 | 1,770 @ 128k-token prompt est. TTFT @8k ≈ 4.6 s (calculated from measured prefill) | ⚡ spec-decode (MTP)context: 128k tokens📏 Same oMLX session as the 1k row. Peak mem 46.4 GB. This generation figure is higher than the 32k cell in the same session; MTP results can vary by prompt. Listed as published. Retrieved 21 Nov 2026. |
| Inkling 2 Mini | oMLX oQ8e, Affine8 8-bit KV | oMLX v0.8.1, Lightning MTP | 55.4 | 1,164 @ 195k-token prompt est. TTFT @8k ≈ 7.0 s (calculated from measured prefill) | ⚡ spec-decode (MTP)context: 195k tokens📏 Same oMLX session as the 1k row. Peak mem 59.4 GB. Retrieved 21 Nov 2026. |
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.
Public oMLX benchmark session, retrieved 21 Nov 2026. Hardware, software, settings and the context-length table are taken from the session pages. Quant is labelled Inkling-2-Mini-oQ8e on oMLX; Lightning MTP and Affine8 8-bit KV were on. This is an independent oMLX run on the same chip class (M5 Ultra, 96 GB) as EXO Labs' vendor-reported 210 tok/s figure, not a re-run of EXO's upcoming engine. RigBase did not re-run it. Session resource panel: peak footprint 64.86 GB, total RAM 96.00 GB.
https://omlx.ai/benchmarks/performance/0z8r79zd
Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.
Quick answers: Can Apple M5 Ultra run Inkling 2 Mini?
Speculative-decoding rows on this page are not like-for-like with plain decode. See how much MTP and DFlash really speed things up (measured pairs).