VeizikNVIDIA, Apple Silicon and edge builds are available now.Release notes →
veiziknative inference engine
For business Sign inDownload
Hardware

Apple M2 Pro

Veizik on Apple M2 Pro — Apple Silicon · Metal · M2.

Overview

What this device is.

Apple M2 ProAvailable
Vendor
Apple Silicon
Backend
Metal
Platform
macOS on Apple Silicon (arm64)
Generation
M2
Architecture
M2
Availability
Available

On Apple M2 Pro, 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
M2
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 familyEvidence on this deviceAvailability
Qwen2.5
Language		Verified VerifiedAvailable
Qwen3
Language		Verified VerifiedAvailable
Performance

Measured on this exact device.

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).

ModelModeWorkloadtok/s
Qwen2.5-0.5B—512 / 128 tok300.6 tok/s · 11 runs · range 299.0–302.5
Qwen2.5-0.5B—512 / 128 tok300.6 tok/s · 11 runs · range 299.0–302.5
Qwen2.5-0.5B—3968 / 128 tok222.0 tok/s · 11 runs · range 221.5–223.4
Qwen2.5-0.5B—3968 / 128 tok222.0 tok/s · 11 runs · range 221.5–223.4
Qwen2.5-0.5B—8192 / 128 tok192.8 tok/s · 11 runs · range 189.2–195.6
Qwen2.5-0.5B—8192 / 128 tok192.8 tok/s · 11 runs · range 189.2–195.6
Qwen2.5-0.5B—16384 / 128 tok141.7 tok/s · 11 runs · range 139.8–143.2
Qwen2.5-0.5B—16384 / 128 tok141.7 tok/s · 11 runs · range 139.8–143.2
Qwen2.5-0.5B—24576 / 128 tok120.9 tok/s · 11 runs · range 120.3–121.3
Qwen2.5-0.5B—24576 / 128 tok120.9 tok/s · 11 runs · range 120.3–121.3
Qwen2.5-0.5B—30000 / 128 tok110.3 tok/s · 11 runs · range 109.8–110.9
Qwen2.5-0.5B—30000 / 128 tok110.3 tok/s · 11 runs · range 109.8–110.9
Qwen2.5-1.5B—512 / 128 tok144.6 tok/s · 11 runs · range 144.3–146.6
Qwen2.5-1.5B—512 / 128 tok144.6 tok/s · 11 runs · range 144.3–146.6
Qwen2.5-1.5B—3968 / 128 tok113.7 tok/s · 11 runs · range 109.6–116.6
Qwen2.5-1.5B—3968 / 128 tok113.7 tok/s · 11 runs · range 109.6–116.6
Qwen2.5-1.5B—8192 / 128 tok102.4 tok/s · 11 runs · range 101.6–103.6
Qwen2.5-1.5B—8192 / 128 tok102.4 tok/s · 11 runs · range 101.6–103.6
Qwen2.5-1.5B—16384 / 128 tok86.3 tok/s · 11 runs · range 85.8–86.5
Qwen2.5-1.5B—16384 / 128 tok86.3 tok/s · 11 runs · range 85.8–86.5
Qwen2.5-1.5B—24576 / 128 tok74.1 tok/s · 11 runs · range 73.8–74.5
Qwen2.5-1.5B—24576 / 128 tok74.1 tok/s · 11 runs · range 73.8–74.5
Qwen2.5-1.5B—30000 / 128 tok67.8 tok/s · 11 runs · range 67.5–68.2
Qwen2.5-1.5B—30000 / 128 tok67.8 tok/s · 11 runs · range 67.5–68.2
Qwen2.5-7B—512 / 128 tok38.3 tok/s · 11 runs · range 38.0–38.4
Qwen2.5-7B—512 / 128 tok38.3 tok/s · 11 runs · range 38.0–38.4
Qwen2.5-7B—3968 / 128 tok33.0 tok/s · 11 runs · range 32.8–33.1
Qwen2.5-7B—3968 / 128 tok33.0 tok/s · 11 runs · range 32.8–33.1
Qwen2.5-7B—8192 / 128 tok32.3 tok/s · 11 runs · range 32.1–32.5
Qwen2.5-7B—8192 / 128 tok32.3 tok/s · 11 runs · range 32.1–32.5
Qwen2.5-7B—16384 / 128 tok28.5 tok/s · 11 runs · range 28.4–28.6
Qwen2.5-7B—16384 / 128 tok28.5 tok/s · 11 runs · range 28.4–28.6
Qwen2.5-7B—24576 / 128 tok25.3 tok/s · 11 runs · range 25.2–25.3
Qwen2.5-7B—24576 / 128 tok25.3 tok/s · 11 runs · range 25.2–25.3
Qwen2.5-7B—30000 / 128 tok23.5 tok/s · 11 runs · range 23.5–23.6
Qwen2.5-7B—30000 / 128 tok23.5 tok/s · 11 runs · range 23.5–23.6
Qwen3-1.7B—512 / 128 tok143.8 tok/s · 11 runs · range 142.5–144.4
Qwen3-1.7B—512 / 128 tok143.8 tok/s · 11 runs · range 142.5–144.4
Qwen3-1.7B—3968 / 128 tok106.6 tok/s · 11 runs · range 106.3–106.8
Qwen3-1.7B—3968 / 128 tok106.6 tok/s · 11 runs · range 106.3–106.8
Qwen3-1.7B—8192 / 128 tok83.9 tok/s · 11 runs · range 83.6–84.2
Qwen3-1.7B—8192 / 128 tok83.9 tok/s · 11 runs · range 83.6–84.2
Qwen3-1.7B—16384 / 128 tok60.1 tok/s · 11 runs · range 60.0–60.2
Qwen3-1.7B—16384 / 128 tok60.1 tok/s · 11 runs · range 60.0–60.2
Qwen3-1.7B—24576 / 128 tok46.4 tok/s · 11 runs · range 46.2–46.4
Qwen3-1.7B—24576 / 128 tok46.4 tok/s · 11 runs · range 46.2–46.4
Qwen3-1.7B—30000 / 128 tok40.3 tok/s · 11 runs · range 40.2–40.3
Qwen3-1.7B—30000 / 128 tok40.3 tok/s · 11 runs · range 40.2–40.3

Prefill, tokens per second Measured here

Prompt tokens processed per second.

ModelModeWorkloadtok/s
Qwen2.5-0.5B—512 / 128 tok5,060 tok/s · 11 runs · range 4,985–5,101
Qwen2.5-0.5B—512 / 128 tok5,060 tok/s · 11 runs · range 4,985–5,101
Qwen2.5-0.5B—3968 / 128 tok4,664 tok/s · 11 runs · range 4,657–4,673
Qwen2.5-0.5B—3968 / 128 tok4,664 tok/s · 11 runs · range 4,657–4,673
Qwen2.5-0.5B—8192 / 128 tok3,917 tok/s · 11 runs · range 3,914–3,919
Qwen2.5-0.5B—8192 / 128 tok3,917 tok/s · 11 runs · range 3,914–3,919
Qwen2.5-0.5B—16384 / 128 tok2,948 tok/s · 11 runs · range 2,946–2,948
Qwen2.5-0.5B—16384 / 128 tok2,948 tok/s · 11 runs · range 2,946–2,948
Qwen2.5-0.5B—24576 / 128 tok2,356 tok/s · 11 runs · range 2,355–2,357
Qwen2.5-0.5B—24576 / 128 tok2,356 tok/s · 11 runs · range 2,355–2,357
Qwen2.5-0.5B—30000 / 128 tok2,071 tok/s · 11 runs · range 2,070–2,072
Qwen2.5-0.5B—30000 / 128 tok2,071 tok/s · 11 runs · range 2,070–2,072
Qwen2.5-1.5B—512 / 128 tok1,506 tok/s · 11 runs · range 1,498–1,511
Qwen2.5-1.5B—512 / 128 tok1,506 tok/s · 11 runs · range 1,498–1,511
Qwen2.5-1.5B—3968 / 128 tok1,448 tok/s · 11 runs · range 1,447–1,448
Qwen2.5-1.5B—3968 / 128 tok1,448 tok/s · 11 runs · range 1,447–1,448
Qwen2.5-1.5B—8192 / 128 tok1,273 tok/s · 11 runs · range 1,272–1,273
Qwen2.5-1.5B—8192 / 128 tok1,273 tok/s · 11 runs · range 1,272–1,273
Qwen2.5-1.5B—16384 / 128 tok1,025 tok/s · 11 runs · range 1,025–1,025
Qwen2.5-1.5B—16384 / 128 tok1,025 tok/s · 11 runs · range 1,025–1,025
Qwen2.5-1.5B—24576 / 128 tok854.5 tok/s · 11 runs · range 854.3–854.7
Qwen2.5-1.5B—24576 / 128 tok854.5 tok/s · 11 runs · range 854.3–854.7
Qwen2.5-1.5B—30000 / 128 tok769.6 tok/s · 11 runs · range 769.5–769.7
Qwen2.5-1.5B—30000 / 128 tok769.6 tok/s · 11 runs · range 769.5–769.7
Qwen2.5-7B—512 / 128 tok333.1 tok/s · 11 runs · range 332.6–333.5
Qwen2.5-7B—512 / 128 tok333.1 tok/s · 11 runs · range 332.6–333.5
Qwen2.5-7B—3968 / 128 tok327.2 tok/s · 11 runs · range 326.9–327.3
Qwen2.5-7B—3968 / 128 tok327.2 tok/s · 11 runs · range 326.9–327.3
Qwen2.5-7B—8192 / 128 tok306.8 tok/s · 11 runs · range 306.8–307.0
Qwen2.5-7B—8192 / 128 tok306.8 tok/s · 11 runs · range 306.8–307.0
Qwen2.5-7B—16384 / 128 tok271.9 tok/s · 11 runs · range 271.9–272.0
Qwen2.5-7B—16384 / 128 tok271.9 tok/s · 11 runs · range 271.9–272.0
Qwen2.5-7B—24576 / 128 tok243.6 tok/s · 11 runs · range 243.6–243.7
Qwen2.5-7B—24576 / 128 tok243.6 tok/s · 11 runs · range 243.6–243.7
Qwen2.5-7B—30000 / 128 tok227.6 tok/s · 11 runs · range 227.6–227.7
Qwen2.5-7B—30000 / 128 tok227.6 tok/s · 11 runs · range 227.6–227.7
Qwen3-1.7B—512 / 128 tok1,459 tok/s · 11 runs · range 1,450–1,461
Qwen3-1.7B—512 / 128 tok1,459 tok/s · 11 runs · range 1,450–1,461
Qwen3-1.7B—3968 / 128 tok1,306 tok/s · 11 runs · range 1,305–1,306
Qwen3-1.7B—3968 / 128 tok1,306 tok/s · 11 runs · range 1,305–1,306
Qwen3-1.7B—8192 / 128 tok1,112 tok/s · 11 runs · range 1,112–1,112
Qwen3-1.7B—8192 / 128 tok1,112 tok/s · 11 runs · range 1,112–1,112
Qwen3-1.7B—16384 / 128 tok880.2 tok/s · 11 runs · range 879.5–880.8
Qwen3-1.7B—16384 / 128 tok880.2 tok/s · 11 runs · range 879.5–880.8
Qwen3-1.7B—24576 / 128 tok727.0 tok/s · 11 runs · range 726.7–727.4
Qwen3-1.7B—24576 / 128 tok727.0 tok/s · 11 runs · range 726.7–727.4
Qwen3-1.7B—30000 / 128 tok651.1 tok/s · 11 runs · range 650.8–651.6
Qwen3-1.7B—30000 / 128 tok651.1 tok/s · 11 runs · range 650.8–651.6

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.

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 familyApple M2 ProState
Qwen2.5VerifiedVerified
Qwen3VerifiedVerified

No sibling device of this architecture is in the matrix.