Builds / Apple M5 Ultra (80-core GPU, unified)
Mac Studio M5 Ultra 256 GB: Qwen3 30B-A3B 162.2 tok/s, gpt-oss 120B 121.4, Llama 3.3 70B 25.2 on MLX, two units within 0.35% (Bohemian Silicon)
Last updated:
- GPU(s)
- Apple M5 Ultra (80-core GPU, unified)
- VRAM / unified memory (GB)
- 256
- CPU
- Apple M5 Ultra (36-core CPU)
- Motherboard
- Mac Studio
Amazon · eBay (used) · Newegg - RAM
- 256 GB unified
- PSU
- not stated
- Other
- macOS 27.2.1, Python 3.12, MLX 0.32.4, mlx-lm 0.32.2; two identical Macs (one in use since October, one new)
- Reported total cost
- not stated
- Power
- not stated
Measured speed (generation tokens/sec)
| Model | Quant | Engine | tok/s | Prompt processing (tok/s) | Details |
|---|---|---|---|---|---|
| gpt-oss 120B | MXFP4-Q8 MLX (mlx-community) | mlx-lm 0.32.2 (MLX 0.32.4) | 121.4 | 5,143 @ 8K-token prompt est. TTFT @8k ≈ 1.6 s (calculated from measured prefill) | context: 512-token prompt, 128 generated📏 Bohemian Silicon run 11 Nov 2026: mlx_lm.benchmark, 512-token random prompt + 128 generated tokens, average of 3 runs (older of two identical Macs; the newer one was within 0.35%); 114.8 tok/s with an 8,192-token prompt; peak memory 63.8 GB |
| Qwen3 30B-A3B legacy | 4-bit MLX (mlx-community) | mlx-lm 0.32.2 (MLX 0.32.4) | 162.2 | 7,646 @ 8K-token prompt est. TTFT @8k ≈ 1.1 s (calculated from measured prefill) | context: 512-token prompt, 128 generated📏 Bohemian Silicon run 11 Nov 2026: mlx_lm.benchmark, 512-token random prompt + 128 generated tokens, average of 3 runs (older of two identical Macs; the newer one was within 0.35%); 138.1 tok/s with an 8,192-token prompt; peak memory 17.8 GB |
| Llama 3.3 70B legacy | 4-bit MLX (mlx-community) | mlx-lm 0.32.2 (MLX 0.32.4) | 25.2 | 682 @ 8K-token prompt est. TTFT @8k ≈ 12.0 s (calculated from measured prefill) | context: 512-token prompt, 128 generated📏 Bohemian Silicon run 11 Nov 2026: mlx_lm.benchmark, 512-token random prompt + 128 generated tokens, average of 3 runs (older of two identical Macs; the newer one was within 0.35%); 23.6 tok/s with an 8,192-token prompt; peak memory 40.3 GB |
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.
Measured by Bohemian Silicon, a Prague company that rents out these Macs (owner-run, published 11 Nov 2026). Rows are the older Mac; the new unit was within 0.35% on generation and 0.7% on long-prompt processing. gpt-oss 120B matched their October run (120 tok/s). Random-token prompts give pure speed figures; real tasks can differ. Same Mac model: the Qwen3.5 397B-A17B build.
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 gpt-oss 120B?