Hardware

Apple M1 Max

Veizik on Apple M1 Max — Apple Silicon · Metal · M1.

Overview

What this device is.

Apple M1 MaxAvailable

Vendor: Apple Silicon
Backend: Metal
Platform: macOS on Apple Silicon (arm64)
Generation: M1
Architecture: M1
Availability: Available

On Apple M1 Max, 2 model families are verified end to end on a published build. Evidence and availability are separate: “verified” is a first-hand run on this exact device, not a download, and “available” never means verified on your hardware.

All devices · Platform requirements · Benchmarks & method

Engines

The engine that runs here.

Veizik native engineAvailable

Backend: Metal
Architecture: M1
Model path: dep0 — pure native, no Python runtime
Families carried: Qwen2.5, Qwen3

One engine runs every model family on this backend. Each GPU architecture is carried as its own native kernel set; a family is “verified” on a device only when it has been run there end to end.

Models

What runs on this device.

Pick a device, get a straight answer — which model families are verified, carried by architecture, or in preparation.

VerifiedVerifiedRun end to end on this exact device, on a published build.Architecture-supportedArchitecture-supportedVerified on another device of the same architecture; this device itself is untested.Expected, untestedExpected, untestedThe engine targets this architecture; nobody has run this model on it. No claim is made.In preparationIn preparationNo build is downloadable yet, and no figure is offered.Not offeredNot offeredThis family is not offered on this platform.UnknownUnknownNo information yet — the platform is announced with the release.

Model family	Evidence on this device	Availability
Qwen2.5 Language	Verified Verified	Available
Qwen3 Language	Verified Verified	Available

Performance

Measured on this exact device.

Model	Task	Weights in memory
Qwen2.5-0.5B	Language	343.6 MB
Qwen2.5-1.5B	Language	926.9 MB
Qwen2.5-7B	Language	4,114.0 MB
Qwen3-1.7B	Language	968.0 MB

Deterministic load — the weights occupy the same memory every run.

Decode, tokens per second Measured here

Generated tokens per second, greedy, one request, through veizik run. Median of the runs shown (hover for the range).

Model	Mode	Workload	tok/s
Qwen2.5-0.5B	—	512 / 128 tok	273.1 tok/s · 11 runs · range 263.5–282.1
Qwen2.5-0.5B	—	3968 / 128 tok	208.9 tok/s · 11 runs · range 207.5–211.6
Qwen2.5-0.5B	—	8192 / 128 tok	209.9 tok/s · 11 runs · range 203.3–212.2
Qwen2.5-0.5B	—	16384 / 128 tok	170.8 tok/s · 11 runs · range 167.6–172.8
Qwen2.5-0.5B	—	24576 / 128 tok	144.1 tok/s · 11 runs · range 142.3–146.1
Qwen2.5-0.5B	—	30000 / 128 tok	154.4 tok/s · 11 runs · range 152.4–161.3
Qwen2.5-1.5B	—	512 / 128 tok	181.1 tok/s · 11 runs · range 179.7–184.9
Qwen2.5-1.5B	—	3968 / 128 tok	155.3 tok/s · 11 runs · range 153.4–158.0
Qwen2.5-1.5B	—	8192 / 128 tok	136.9 tok/s · 11 runs · range 136.1–137.9
Qwen2.5-1.5B	—	16384 / 128 tok	119.6 tok/s · 11 runs · range 118.5–121.0
Qwen2.5-1.5B	—	24576 / 128 tok	99.3 tok/s · 11 runs · range 98.9–100.9
Qwen2.5-1.5B	—	30000 / 128 tok	91.5 tok/s · 11 runs · range 90.7–92.2
Qwen2.5-7B	—	512 / 128 tok	60.4 tok/s · 11 runs · range 59.5–60.5
Qwen2.5-7B	—	3968 / 128 tok	51.7 tok/s · 11 runs · range 50.6–51.8
Qwen2.5-7B	—	8192 / 128 tok	50.5 tok/s · 11 runs · range 49.9–50.6
Qwen2.5-7B	—	16384 / 128 tok	44.3 tok/s · 11 runs · range 44.1–44.5
Qwen2.5-7B	—	24576 / 128 tok	39.4 tok/s · 11 runs · range 39.1–39.6
Qwen2.5-7B	—	30000 / 128 tok	36.8 tok/s · 11 runs · range 36.3–36.9
Qwen3-1.7B	—	512 / 128 tok	207.4 tok/s · 11 runs · range 206.2–209.4
Qwen3-1.7B	—	3968 / 128 tok	163.2 tok/s · 11 runs · range 162.2–164.7
Qwen3-1.7B	—	8192 / 128 tok	127.8 tok/s · 11 runs · range 124.6–128.1
Qwen3-1.7B	—	16384 / 128 tok	96.6 tok/s · 11 runs · range 94.9–97.0
Qwen3-1.7B	—	24576 / 128 tok	76.8 tok/s · 11 runs · range 74.8–77.0
Qwen3-1.7B	—	30000 / 128 tok	67.5 tok/s · 11 runs · range 65.8–67.8

Prefill, tokens per second Measured here

Prompt tokens processed per second.

Model	Mode	Workload	tok/s
Qwen2.5-0.5B	—	512 / 128 tok	7,901 tok/s · 11 runs · range 6,417–8,103
Qwen2.5-0.5B	—	3968 / 128 tok	7,504 tok/s · 11 runs · range 7,428–7,555
Qwen2.5-0.5B	—	8192 / 128 tok	6,381 tok/s · 11 runs · range 6,362–6,411
Qwen2.5-0.5B	—	16384 / 128 tok	4,829 tok/s · 11 runs · range 4,822–4,838
Qwen2.5-0.5B	—	24576 / 128 tok	3,859 tok/s · 11 runs · range 3,853–3,862
Qwen2.5-0.5B	—	30000 / 128 tok	3,434 tok/s · 11 runs · range 3,428–3,546
Qwen2.5-1.5B	—	512 / 128 tok	2,440 tok/s · 11 runs · range 2,428–2,443
Qwen2.5-1.5B	—	3968 / 128 tok	2,443 tok/s · 11 runs · range 2,441–2,444
Qwen2.5-1.5B	—	8192 / 128 tok	2,217 tok/s · 11 runs · range 2,214–2,218
Qwen2.5-1.5B	—	16384 / 128 tok	1,831 tok/s · 11 runs · range 1,831–1,831
Qwen2.5-1.5B	—	24576 / 128 tok	1,554 tok/s · 11 runs · range 1,554–1,554
Qwen2.5-1.5B	—	30000 / 128 tok	1,411 tok/s · 11 runs · range 1,411–1,411
Qwen2.5-7B	—	512 / 128 tok	545.9 tok/s · 11 runs · range 545.6–546.3
Qwen2.5-7B	—	3968 / 128 tok	552.0 tok/s · 11 runs · range 551.8–552.2
Qwen2.5-7B	—	8192 / 128 tok	520.9 tok/s · 11 runs · range 517.5–521.1
Qwen2.5-7B	—	16384 / 128 tok	465.1 tok/s · 11 runs · range 465.0–465.5
Qwen2.5-7B	—	24576 / 128 tok	419.2 tok/s · 11 runs · range 419.1–419.3
Qwen2.5-7B	—	30000 / 128 tok	392.6 tok/s · 11 runs · range 386.4–392.7
Qwen3-1.7B	—	512 / 128 tok	2,417 tok/s · 11 runs · range 2,394–2,421
Qwen3-1.7B	—	3968 / 128 tok	2,233 tok/s · 11 runs · range 2,231–2,234
Qwen3-1.7B	—	8192 / 128 tok	1,958 tok/s · 11 runs · range 1,955–1,960
Qwen3-1.7B	—	16384 / 128 tok	1,562 tok/s · 11 runs · range 1,561–1,563
Qwen3-1.7B	—	24576 / 128 tok	1,296 tok/s · 11 runs · range 1,295–1,296
Qwen3-1.7B	—	30000 / 128 tok	1,156 tok/s · 11 runs · range 1,152–1,158

Every figure is veizik’s own measurement on this exact device, bound to the published build it was taken on. See Reproduce our benchmarks.

Compatibility

Architecture and its siblings.

Model family	Apple M1 Max	State
Qwen2.5	Verified	Verified
Qwen3	Verified	Verified

Same architecture (M1), so a family verified on one carries to the others as Supported: Apple M1 Ultra.