VeizikNVIDIA and Apple Silicon builds are available now.Release notes →
veiziknative inference engine
Sign inDownload
Product
OverviewWhat the engine isHow it worksWeights in, native execution outPlatformsNVIDIA · Apple SiliconSecurity & licensingDevice-bound, local by designReleasesEvery build, dated
Engineering
How it worksOne researched page per capabilityVideo memory720p that stays under a 24 GB cardA 35B model on one card21 GiB resident on a 24 GB GPUIts own stackNo Python, no framework, no cuBLASResume a renderCrash, resume, same result
Models
NVIDIALanguage and video — available nowApple SiliconQwen2.5 · Qwen3 — availableFull catalogueEvery model and its status
Hardware
All hardwareEvery device × every modelPick your deviceWhat runs on yoursNVIDIAAmpere → Blackwell · CUDAApple SiliconM1 – M4 · Metal
Benchmarks
All resultsMatrix or pick, by buildBuilds & downloadsEvery build, JSON / CSVMethod & notationHow every figure is taken
Docs
QuickstartInstall to first runCommand referenceEvery documented commandActivate a machineDevices and licenceWhat leaves the machineOffline and networkTroubleshootingWhen something fails
ReleasesSign inDownloadGitHub
Hardware

What runs on your hardware.

Every device we build for, every model family, and the evidence behind each cell. Show everything, or pick your device.

Show all

Devices × model families.

Grouped by vendor and generation, with the architecture code beside each device.

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.
DeviceFLUX.1Stable Diffusion 3.5Wan 2.1 · Wan 2.2Qwen3.5Qwen3.6
NVIDIA · CUDA · Linux on x86-64 Available Target details →
Ampere
NVIDIA GeForce RTX 3080 Ti sm_86In preparationIn preparationArchitecture-supportedArchitecture-supportedArchitecture-supported
GeForce RTX 3090 sm_86In preparationIn preparationVerifiedVerifiedVerified
NVIDIA RTX A5000 sm_86In preparationIn preparationArchitecture-supportedVerifiedArchitecture-supported
NVIDIA A40 sm_86In preparationIn preparationArchitecture-supportedArchitecture-supportedArchitecture-supported
NVIDIA A100 80GB PCIe sm_80In preparationIn preparationExpected, untestedExpected, untestedExpected, untested
NVIDIA A100-SXM4-80GB sm_80In preparationIn preparationExpected, untestedExpected, untestedExpected, untested
Ada Lovelace
NVIDIA GeForce RTX 4090 sm_89In preparationIn preparationExpected, untestedVerifiedVerified
Hopper
NVIDIA H100 80GB HBM3 sm_90In preparationIn preparationExpected, untestedExpected, untestedExpected, untested
Blackwell
NVIDIA GeForce RTX 5090 sm_120In preparationIn preparationExpected, untestedVerifiedVerified
NVIDIA RTX PRO 4500 Blackwell sm_120In preparationIn preparationExpected, untestedArchitecture-supportedArchitecture-supported
DeviceQwen2.5Qwen3
Apple Silicon · Metal · macOS on Apple Silicon (arm64) Available Target details →
M1
Apple M1 Ultra M1VerifiedVerified
Apple M1 Max M1VerifiedVerified
M2
Apple M2 Pro M2VerifiedVerified
M4
Apple M4 Pro M4VerifiedVerified

Evidence and availability are separate. “Verified” never means downloadable, and “Available” never means verified on your device. Data generated 2026-10-28. Families the engine is compatible with but that are not offered are listed in the models catalogue.

Stable Diffusion 3.5: Powered by Stability AI

Pick your device

Choose a device, get a straight answer.

Requirements

Per platform.

NVIDIA
Platform
Linux on x86-64
Backend
CUDA
Generations
Ampere, Ada Lovelace, Hopper, Blackwell
Minimums
An NVIDIA GPU of compute capability 8.0 or newer, and an NVIDIA driver that provides libcuda.so.1. No CUDA toolkit and no Python — only the driver. The same build runs under both the CUDA 12 and CUDA 13 driver series; there is no separate CUDA-12 or CUDA-13 download. Linux x86-64, run natively (WSL is not supported).
Apple Silicon
Platform
macOS on Apple Silicon (arm64)
Backend
Metal
Generations
M1, M2, M3, M4
Minimums
Apple Silicon (M1 or newer), and macOS 13 or later. It is not tied to one macOS version: 13 is the minimum tested, and newer releases run.

No bundled dependencies. The engine process is a single binary, and it links only three shared libraries: the NVIDIA driver (libcuda.so.1), the C library (libc) and the math library (libm). No CUDA runtime, no cuBLAS or cuDNN, no PyTorch and no Python — not installed, and not hidden inside the package. The package also carries a small launcher and a seal checker; those link only libc. On Apple Silicon the one dependency is Metal. You can check this yourself on the files you download, with readelf -d (or ldd) for the linked libraries, sha256sum -c against the published checksums, and the bundled seal checker.

Verify it yourself

No trust required — three commands on the files you download:

$ sha256sum veizik-<model>-<version>-linux-x86_64.tar   # compare to the SHA-256 on the release page
$ bin/vzseal <unpacked-folder>                          # checks every sealed file — prints VZSEAL OK
$ readelf -d bin/<engine> | grep NEEDED                  # or ldd — lists the libraries it links

The third command prints the dependency list above, straight from the binary. To reproduce the measured figures as well, follow Reproduce our benchmarks.