llamaperf

Mac M6 LLM benchmarks

Every M6 configuration on llamaperf, M6, with its unified memory, bandwidth and the generation and prompt-processing speeds people report. Memory decides which models fit; bandwidth shapes both speeds.

3 configurations · 23 reports

As of 22 Nov 2026, the models most run on M6 Macs, each on one configuration, with the median of plain runs (one machine, one request, no speculative decoding):

Configurations

ChipMemoryBandwidthGen / PP t/sReports
M632 GB171 GB/s6422
M624 GB171 GB/s1
M616 GB154 GB/s0

Generation and PP medians are plain single-machine runs of Qwen3.6 35B · 3B active at 4-bit, so every row is the same model; a blank means no run of it on that configuration yet. 1 configuration has no report yet.

Models people run on M6 Macs

Latest reports

Showing the 20 newest of 23 reports. See all of them →

Speeds are as each source reported them. GPU count, offloading and concurrent requests change them, and prompt-processing speed is input throughput, not time to first token. Check the full setup before comparing two reports.

Qwen3.8 27B

M6 32GB · ANEMLL Forge · 96,000 ctx

Tone: positiveApple hardware
generation:
27.0 tokens/s
prompt processing (prefill):
123.0 tokens/s
quant:
INT8
kv:
INT8

User reports Qwen3.8-27B running on the M6 Neural Engine via ANEMLL Forge 0.2.1, with about 27 tok/s decode and 123 tok/s prefill at 96K context on a 32GB M6 Mac Mini. Setup uses an INT8 K/V cache and a smaller drafter sharing the target's output head, with about 21.4 GB total memory at 96K context. Decode varies by content: JSON 35–62 tok/s, code 18–32, prose 10–17. Prefill is 260 tok/s at short context and 62 at 192K, where a smaller prefill batch of 32 rows is used. The user calls 96K the sweet spot for size and speed on 32GB.

Oct 24, 2026
Tone: positiveApple hardware
generation:
10.0 tokens/s
prompt processing (prefill):
200.0 tokens/s
quant:
Q2_0 (GGUF)

User reports Qwen3.8-Flash-Next at about 10 tok/s generation and about 200 tok/s prefill on a Mac Mini M6 with 32 GB. Setup is Strata with a Q2_0 quant at 64K context. User says quality seems on par with 4-bit Qwen3.8-27B and that the fans almost never turn on with Strata, unlike the 27B run with Splash.

Oct 23, 2026

Qwen3.8 27B

M6 32GB · Inco Splash

Tone: positiveApple hardware
generation:
33-35 tokens/s

User reports Qwen3.8 27B running at 33-35 t/s on an M6 32GB Mac using the Inco Splash inference engine. Other engines topped out at 9.6 t/s on the same machine. The user is considering an M5 Pro for its 1.7x memory bandwidth and hopes Inco AI adds support for more models like Flash-Next.

Oct 14, 2026

Gemma 3 27B

M6 32GB · MLX-LM 0.31.3 / MLX 0.32.2 / Metal

Apple hardware
generation:
8.9 tokens/s
prompt processing (prefill):
11.2 tokens/s
quant:
4-bit (MLX)

User reports Gemma 3 27B at 8.9 t/s generation and 11.2 t/s prompt processing on a Mac mini M6 with 32GB unified memory. Setup is MLX-LM with the mlx-community 4-bit quantized model, run through the mlx_lm.generate CLI with a 20-token prompt and 500 output tokens. Peak memory was 16.296 GB. The rates are native MLX-LM phase timings from a single displayed run, not a benchmark average.

Oct 13, 2026

Qwen3.5 9B

M6 32GB · LM Studio · 165,120 ctx

Apple hardware
generation:
30.3 tokens/s

User reports Qwen3.5 9B at 30.3 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with the MLX backend and a loaded context of 165,120 tokens, one machine with one model loaded alone. The 30.3 t/s is the median of three English-prose runs, a client-observed streaming decode rate rather than a pooled or peak figure; exact quantization is unknown.

Oct 13, 2026

Gemma 4 31B

M6 32GB · LM Studio · 32,768 ctx

Apple hardware
generation:
7.5 tokens/s

User reports Gemma 4 31B at 7.5 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with a 32,768-token loaded context; quantization and backend are unrecorded. The 7.5 t/s is the median of three English-prose runs, measured as client-observed streaming decode throughput rather than a pooled or peak rate.

Oct 13, 2026

Gemma 4 12B

M6 32GB · LM Studio · 32,768 ctx

Apple hardware
generation:
18.8 tokens/s

User reports Gemma 4 12B at 18.8 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with a QAT-labeled model at 32,768 tokens of loaded context; exact quantization and backend are unknown. The 18.8 t/s is the median of three English-prose runs, client-observed streaming decode throughput, not a pooled rate or prompt-processing rate.

Oct 13, 2026

Gemma 4 26B (4B active)

M6 32GB · LM Studio · 80,128 ctx

Apple hardware
generation:
48.2 tokens/s

User reports Gemma 4 26B-A4B at 48.2 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with an unreported backend and unknown quantization, loaded context 80,128 tokens, one model loaded alone. The 48.2 t/s is the median of three English-prose runs, client-observed streaming decode throughput, not a pooled rate or peak; the five workload medians range from 45.5 to 53.3 t/s.

Oct 13, 2026

gpt-oss 21B (3.6B active)

M6 32GB · LM Studio · 32,768 ctx

Apple hardware
generation:
45.2 tokens/s

User reports gpt-oss 21B-A3.6B at 45.2 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with an unreported backend and quantization, loaded at 32,768 tokens of context on one machine. The 45.2 t/s is the median of three English-prose runs and is client-observed streaming decode throughput, not a pooled rate or prompt-processing rate; the JSON extraction task reached a 53.3 t/s median.

Oct 13, 2026

Muse Glimmer

M6 32GB · LM Studio · 32,768 ctx

Apple hardware
generation:
8.6 tokens/s

User reports Muse at 8.6 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with an unknown quantization and a loaded context of 32,768 tokens on one machine. The 8.6 t/s is the median of three English-prose runs, client-observed streaming decode throughput rather than a pooled or peak rate; the JSON task passed 0 of 3 automatic checks.

Oct 13, 2026

Nemotron 3 Nano Omni

M6 32GB · LM Studio · 32,768 ctx

Apple hardware
generation:
45.0 tokens/s

User reports Nemotron 3 at 45 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with an unknown quantization and a loaded context of 32,768 tokens on one machine. The 45 t/s is the median of three English-prose runs and is client-observed streaming decode throughput, not a pooled rate or prompt-processing rate.

Oct 13, 2026

Ornith1.5 35B (3B active) Splash

M6 32GB · LM Studio · 32,768 ctx

Apple hardware
generation:
61.5 tokens/s

User reports Ornith1.5-35B-A3B Splash at 61.5 t/s on an Apple Mac mini M6 with 32GB unified memory. Setup is LM Studio with the Splash backend and 32,768 tokens of loaded context, one model loaded alone. The 61.5 t/s is the median of three English-prose runs, client-observed streaming decode throughput rather than a pooled rate or peak; the three runs measured 64.7, 59.4 and 61.5 t/s.

Oct 13, 2026

Qwen3.8 27B Splash

M6 32GB · LM Studio · 32,768 ctx

Apple hardware
generation:
20.2 tokens/s

User reports Qwen3.8-27B Splash at 20.2 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with the Splash backend, 32,768 tokens of loaded context, and speculative decoding with DFlash2 drafts; quantization is unknown. The 20.2 t/s is the median of three English-prose runs, with individual runs of 20.6, 20.2 and 20 t/s; it is client-observed streaming decode throughput, not a pooled rate or prompt-processing rate.

Oct 13, 2026

Qwen3.8 27B Swift Splash

M6 32GB · LM Studio · 32,768 ctx

Apple hardware
generation:
19.5 tokens/s

User reports Swift-Qwen3.8-27B Splash at 19.5 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with the Splash backend and speculative decoding using DFlash2 drafts, loaded at 32,768 tokens of context. The 19.5 t/s is the median of three English-prose runs and is client-observed streaming decode throughput, not a pooled rate or prompt-processing rate; exact quantization is unknown.

Oct 13, 2026

Qwen3.8 27B

M6 32GB · LM Studio · 32,768 ctx

Apple hardware
generation:
11.8 tokens/s
quant:
Q4_K_M (GGUF)

User reports Qwen3.8 27B at 11.8 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with its llama.cpp GGUF backend, GGUF Q4_K_M quantization, and a loaded context of 32,768 tokens. The 11.8 t/s is the median of three English-prose runs (Ruhr economy prompt), client-observed streaming decode throughput, not an overall mean or prefill rate.

Oct 13, 2026

Ornith1.5 35B (3B active)

M6 32GB · LM Studio · 32,768 ctx

Apple hardware
generation:
43.7 tokens/s
quant:
Q4_K_M (GGUF)

User reports Ornith1.5 35B-A3B at 43.7 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with its llama.cpp GGUF backend, GGUF Q4_K_M, and a loaded context of 32,768 tokens. The 43.7 t/s is the median of three English-prose runs (Ruhr economy prompt) and is client-observed streaming decode throughput, not an overall mean, peak speed, or prefill rate.

Oct 13, 2026
Apple hardware
generation:
42.3 tokens/s
quant:
UD-Q3_K_XL (GGUF)

User reports Nemotron 3.5 Lightning 30B-A3B at 42.3 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with its llama.cpp GGUF backend, UD-Q3_K_XL quant, and 32,768 tokens of loaded context. The 42.3 t/s is the median of three English-prose runs (Ruhr economy prompt), client-observed streaming decode throughput, not an overall mean or prefill rate.

Oct 13, 2026
Apple hardware
generation:
45.1 tokens/s
quant:
Q4_K_M (GGUF)

User reports Nemotron 3.5 Lightning 30B-A3B at 45.1 t/s on an Apple Mac mini M6 with 32 GB unified memory. Setup is LM Studio with its llama.cpp GGUF backend, GGUF Q4_K_M, and a loaded context of 32,768 tokens. The 45.1 t/s is the median of three English-prose runs and is client-observed streaming decode throughput, not an overall mean, peak, or prefill rate.

Oct 13, 2026
Apple hardware
generation:
9.8 tokens/s
quant:
4bit (MLX)
mtp (multi-token prediction):
off
summarization

User reports Qwen3.8 27B at 9.83 t/s generation on a Mac mini M6 24GB, from a 525-token prompt. Setup is mlx-vlm 0.7.3 with the mlx-community 4-bit checkpoint, one request, speculative decoding off. This is the plain baseline from a single run in the author's MTP comparison. The 268 t/s prefill in the source is the prompt divided by time to first token, so no prompt speed is stored.

Oct 9, 2026

Qwen3.5 9B

M6 32GB · llama.cpp

Apple hardware
generation:
27.0 tokens/s
quant:
Q4_K_M (GGUF)

User reports Qwen3.5-9B at approximately 27 tokens/second decode on an Apple Mac mini M6 with 32GB unified memory. Setup is llama.cpp with Q4_K_M weights (5.5GB file) at empty prior context. The figure is from a Tweakers review sweep of prior context from empty to 90,000 tokens in 10,000-token increments; the sweep endpoint is not a verified maximum context capacity. Prefill, KV cache, offload and speculation settings are not reported for this result.

Oct 9, 2026

Frequently asked

Which M6 Mac is best for local LLMs?

On Qwen3.6 35B · 3B active at 4-bit, the M6 model the most of these Macs have been measured on, the M6 32GB is the fastest reported, at 63.8 tokens per second (one run). Speed follows the chip tier's memory bandwidth, and which models fit follows the memory size.

How much unified memory do I need on an M6 Mac for local LLMs?

Enough to hold the model's weights at your quant plus its context, with headroom for macOS. Unified memory is shared with the system, so the usable pool is smaller than the number on the box. The VRAM calculator sizes any model against each configuration.

Why is an M6 Mac slower than an NVIDIA card with less memory?

Generation speed is limited by memory bandwidth, not by memory size. M6 configurations on llamaperf range from about 154 to 171 GB/s, while a high-end discrete GPU reads its smaller pool much faster. A Mac wins on which models fit; a discrete card wins on tokens per second for models that fit both.