M2 Ultra 128GB

APPLE · 128GB unified memory · 4 reports

On the M2 Ultra 128GB, Qwen3-Coder 30B · 3B active at 4-bit, the model most run on it, generates a median of 70.3 tokens per second from 2 plain runs (community reports, as of 30 Oct 2026).

As of 30 Oct 2026, the models most run on the M2 Ultra 128GB, with generation and prompt-processing medians from plain runs (one device, one request, no speculative decoding):

Qwen3-Coder 30B · 3B active
70.3gen t/s at 4-bit · 2 runs
PP 1088.7 t/s · 2 runs

Engines people use on it: llama.cpp 2 (reports) · oMLX 1 (reports)

See reported performance Estimate what fits

How to compare these reports →Filter reports by setup →All M2 Macs compared →

Run models on your M2 Ultra 128GB? Add your numbers.

Your coding agent can run a speed test on your machine and send it here with a free account. Or paste a result you've already posted.

Benchmark with your agent Paste a result

Reported performance on the M2 Ultra 128GB

These reports mix GPU counts and memory setups, so a model listed here may have needed extra cards or CPU offloading. The calculator estimates which models fit in 128 GB of unified memory.

Date ↓Date ↑Token speed ↓Token speed ↑Prefill ↓Prefill ↑Context ↓Context ↑Ranked

Qwen3-Coder 30B (3B active)

M2 Ultra 128GB · llama.cpp

generation:: 71.2 tokens/s
prompt processing (prefill):: 1096.8 tokens/s
quant:: UD-Q4_K_XL (GGUF)
kv:: F16
flash attention:: on
mtp (multi-token prediction):: off

Reported by the source; GPU count, offloading and concurrent requests can change these figures. Prompt-processing speed is input throughput, not time to first token. Check the full setup before comparing.

User reports Qwen3-Coder 30B-A3B at 71.2 t/s generation and 1096.8 t/s prompt processing on a Mac Studio M2 Ultra with 128GB unified memory. Setup is llama.cpp build 6060 with the UD-Q4_K_XL GGUF and flash attention enabled, run with -ngl 100 on one 60-core GPU. The figures are separate synthetic llama-bench tests: PP512 at 1096.84 +/- 5.48 t/s and TG128 at 71.23 +/- 0.08 t/s, with 16 CPU threads and a 16.45 GiB model footprint.

Qwen3-Coder 30B (3B active)

M2 Ultra 128GB · llama.cpp

generation:: 69.3 tokens/s
prompt processing (prefill):: 1080.5 tokens/s
quant:: UD-Q4_K_XL (GGUF)
kv:: F16
flash attention:: off
mtp (multi-token prediction):: off

User reports Qwen3-Coder 30B-A3B at 69.3 t/s generation and 1080.5 t/s prompt processing on a Mac Studio M2 Ultra with 128GB unified memory. Setup is llama.cpp build 6060 with the UD-Q4_K_XL GGUF and flash attention off, one GPU with -ngl 100 and 16 threads. The figures are separate synthetic llama-bench tests: PP512 and TG128, not a 512-input/128-output request. The same run measured 818.85 t/s PP8192 and 40.35 t/s TG8192.

Qwen3.8 27B

M2 Ultra 128GB · oMLX 0.6.3

generation:: 26.4 tokens/s
prompt processing (prefill):: 276.2 tokens/s
quant:: oQ8e mixed affine 8-bit + FP16 (MLX)

User reports Qwen3.8 27B at 26.4 t/s generation and 276.2 t/s prefill on an Apple Mac Studio M2 Ultra with 128 GB unified memory. Setup is oMLX 0.6.3 with oQ8e mixed affine 8-bit plus FP16 weights, run as an ANE and CPU split at a source-labelled 65,536-token prompt depth. The GPU-only control at the same depth is 231.2 prefill and 26.6 decode; the author reports gains apply to cold prefills, not prefix-cached turns, and peak memory at 64K rises from 44.0 GB to 60.0 GB with ANE.

Qwen3.8 27B

M2 Ultra 128GB · oMLX

generation:: 23.0 tokens/s
prompt processing (prefill):: 304.2 tokens/s
quant:: oQ4e (MLX)

summarizationagenticlong-context

User reports Qwen3.8 27B at 23 t/s generation and 304.2 t/s prompt processing on an M2 Ultra 128GB Mac Studio. Setup is oMLX 0.6.3 with oQ4e 4-bit quantization at workload depth 65536, with 16.7GiB of weights and peak memory rising from 25.3 to 36.2GB. The 304.2 PP is the ANE-on arm; the separate GPU-only baseline is 242.2 PP and 23.8 TG, and the source calls the decode change within variance. Depth 65536 is not a verified configured maximum context.

Get a weekly email of new M2 Ultra 128GB reports and newly released models that fit it.

Email me new reports

Reported configurations

Individual observations from this page, not expected speeds or a ranking. Select a model to open its report, with the source and offloading conditions. Reported context may be a limit rather than actual input length.

Model / hardware	Quant / engine	Reported context	Generation
Qwen3-Coder 30B (3B active) M2 Ultra 128GB	UD-Q4_K_XL llama.cpp	Not reported	71.2 tokens/s
Qwen3-Coder 30B (3B active) M2 Ultra 128GB	UD-Q4_K_XL llama.cpp	Not reported	69.3 tokens/s
Qwen3.8 27B M2 Ultra 128GB	oQ8e mixed affine 8-bit + FP16 oMLX 0.6.3	Not reported	26.4 tokens/s
Qwen3.8 27B M2 Ultra 128GB	oQ4e oMLX	Not reported	23.0 tokens/s

Frequently asked

How fast is the M2 Ultra 128GB for local LLMs?

On the M2 Ultra 128GB, Qwen3-Coder 30B · 3B active at 4-bit, the model most run on it, generates a median of 70.3 tokens per second from 2 plain runs (community reports, as of 30 Oct 2026).