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

Choose your build. Run your first model.

Select your platform to see available packages, requirements and the next steps.

Decorative render by the veizik engine.
AI-generated · veizik engine
Reproduce this image

a sleek vertical obsidian monolith, a sealed glossy dark capsule, strong warm amber and rose rim light raking along its edges making the object clearly visible and three-dimensional, a thin glowing rose seam of light down the center, faint softly lit circuit etchings, dramatic single object studio lighting, premium polished product render, dark charcoal #0b0d10 background, deep rose (#da5b75) and warm amber (#e0a458) accent lighting, cinematic volumetric glow, premium, minimal, abstract conceptual, no people, no text, no logos, no watermark, 16:9 wide cinematic composition.

FLUX.1-schnell · seed 555 · 4 steps · 1024x1024 native engine render (edge geometry-lock), center-cropped to 16:9 then upscaled (1600x900 / 800x450)

Rendered by veizik’s own FLUX.1-schnell engine (Apache-2.0) on a Jetson AGX Orin. Decorative only — not a benchmark. The file carries an IPTC AI-generated disclosure.

Platform

Select your platform.

Builds, sizes and licence terms are visible without signing in. Activating a device in step 2 uses your account.

Package · 4 on offer

One veizik command installs any of them by name; each package is its own signed file.

  1. Get VEIZIK

    Before you start This is a 56.7 MB download. veizik install fetches it from its GitHub release and refuses it unless its SHA-256 matches the digest under Verification details. The first install also fetches the small shared-core file.

    bash · Linux on x86-64
    # the veizik command for Linux x86-64
$ curl -fsSL https://veizik.com/install.sh | sh
# if the installer asks you to put ~/.local/bin on your PATH
$ export PATH="$HOME/.local/bin:$PATH"
# accepted names: qwen35, qwen36 or a full model key (qwen35_0p8b, qwen35_2b, qwen35_4b, qwen35_9b, qwen36_35b_a3b); any case; '.' and '-' work too (qwen3.6)
$ veizik install qwen36

  2. Activate this device

    Before you start Activation ties this device to your account and needs a network connection, once per device. Approve the printed address and code in any browser, or on a server use a key and CLI password from your account page.

    activate
    # activation needs no weights; point VZ_WEIGHTS at the checkpoint you downloaded, for veizik run
$ export VZ_WEIGHTS=/data/models
$ veizik activate
# without a browser (SSH, servers): create a key and a CLI password on your account page, then
$ veizik activate --key KEY --password-stdin < pwfile

  3. Run a supported model

    Before you start Model weights are not part of the package. The commands show how each model’s weights are supplied — veizik pull for some, a folder you point to with --weights or VZ_WEIGHTS for others. Weights are far larger than the package, so check free disk space first.

    bash · Qwen3.5
    $ veizik run qwen35_2b "The capital of France is" --tokens 32
$ veizik chat qwen35_2b
$ veizik serve qwen35_2b --weights /data/models/Qwen3.5-2B --port 8000
# then POST http://127.0.0.1:8000/v1/chat/completions
    bash · Qwen3.6
    $ veizik run qwen36_35b_a3b "The capital of France is" --tokens 32
$ veizik chat qwen36_35b_a3b
# image or video input, on Qwen3.6-35B-A3B only
$ veizik run qwen36_35b_a3b "What is in this picture?" --image photo.png --out ./out
$ veizik run qwen36_35b_a3b "What happens in this clip?" --video ./frames --video-fps 2 --out ./out

Requirements

Platform
Linux on x86-64
Minimum
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).
GPU architectures
sm_80, sm_86, sm_89, sm_90a, sm_100a, sm_120a
Hardware generations
Ampere, Ada Lovelace, Hopper, Blackwell
Package download
Qwen3.5 · Qwen3.6: 56.7 MB
Stable Diffusion 3.5: 35.7 MB
Wan 2.1 · Wan 2.2: 7.2 MB
FLUX.1: 45.3 MB

Licence summary

Install and evaluation are free; production use is licensed (see the EULA). Model weights are licensed by their publishers.

This summary is visible without signing in.

Licensing details
Environment notes
  • An NVIDIA GPU with compute capability 8.0 or newer. Both packages carry GPU code for sm_80 and sm_86 (Ampere), sm_89 (Ada Lovelace), sm_90a (Hopper), sm_100a and sm_120a (Blackwell). Qwen package: one GPU runs one model; splitting a model across GPUs is not supported. Qwen package: the image and video input of its Qwen3.6-35B-A3B engine runs on RTX 30-series (sm_86), RTX 40-series (sm_89) and RTX 50-series (sm_120a) class GPUs; the text models keep all six architectures.
  • A driver that provides libcuda.so.1 and supports your card. No CUDA toolkit and no Python are needed. Qwen package: validation used driver 595.91.07; no minimum version is stated.
  • Qwen package: ordinary Linux host tools on the PATH: bash, curl, perl, xxd, od, base64, sed, awk, grep, timeout, mktemp, mkfifo, tee, sha256sum and nvidia-smi.
  • Disk: about 59 MB for the Qwen package and about 7 MB for the Wan package download. The checkpoints you supply are far larger (Qwen3.6-35B-A3B is about 70 GB).
  • Qwen package: network access for activation and licence renewal; generation runs on your GPU and makes no network calls. Wan package: its GPU code opens only on a device activated for this release. Run veizik activate once per device after veizik install wan22. Activation does not need the model weights; rendering does.
  • Qwen package: model weights are not included; download the public Hugging Face checkpoint of each model you want from its publisher. Weights are needed only to run a model — activation does not load or need them. Wan package: model weights are not included. veizik pull wan22 downloads the weights (Apache-2.0) and prepares the video decoder automatically — no manual VAE setup or support request.
Known limits of this build
  • Model weights are not included; the runtime distributes none, and the checkpoints are licensed by their publishers. Qwen package: download them from their publishers; the CLI cannot fetch them. Wan package: veizik pull wan22 downloads the weights (Apache-2.0) and prepares the video decoder automatically — no manual VAE setup or support request.
  • Qwen package, not in this release: LoRA adapters, image or video input on models other than Qwen3.6-35B-A3B, tool or function calling, logprobs and embeddings over HTTP, and a weight converter.
  • Qwen package: image and video input is a capability of its Qwen3.6-35B-A3B engine only: one image (PNG or JPEG) or one video (2 to 256 PNG or JPEG frames, all the same size) per run, with a text prompt, on RTX 30-series (sm_86), RTX 40-series (sm_89) and RTX 50-series (sm_120a) class GPUs only.
  • Qwen package: the server is loopback-only (127.0.0.1); a non-loopback address is refused. To serve other machines, put a reverse proxy that terminates TLS in front of it. Without --require-auth any local process can use it, and authentication is one shared key per server process, not per-user keys.
  • Qwen package: the server has no usage ledger and no per-tenant quotas: --usage-log and --quota-* are accepted but not applied, and so is --max-tokens (set max_tokens in each request). veizik stop does not see the server.
  • Qwen package: the server context window is 4,096 tokens per request (prompt plus generated) in the validated configuration.
  • Qwen package: sampling (--seed, --temperature, --top-p, --top-k) is opt-in and experimental; greedy decoding is the default and the verified path.
  • Qwen package: Qwen3.6-35B-A3B in quality mode needs a GPU with about 26 GB or more and did not complete on a 32 GB card in validation; use balanced or fast. It was not validated for serving on 24 GB cards.
  • Qwen package: GPU code for sm_80, sm_90a and sm_100a is included but was not run in validation.
  • Qwen package: activation needs the network to register the device and install its licence; it runs no model and needs no weights. If the activation flow does not complete — the server declines the machine or the browser approval is not finished — the device is reported as not activated (exit 6) until you run it again.
  • Wan package: every speed figure for it was measured on an NVIDIA RTX 3090 (sm_86) only; other GPUs run it at speeds we have not measured.
  • Wan package: the image-to-video model takes its input image as raw RGB bytes, one byte per channel, at exactly the render size (width × height × 3 bytes); resize it first.
  • Wan package: quality is the reference path and the default; balanced, fast and ultrafast are faster and not bit-identical to it. Without --det-vae yes the decoded frames are not guaranteed to repeat byte for byte.
  • Wan package: the transformers and the text encoder are read as published (F32 safetensors); a BF16 shard is refused.
  • The GPU code is stored sealed, and the key to run it arrives only through device-bound activation; the licence is bound to the device and can be revoked. On Linux, root can read a running process's memory.
Verification

Check what you downloaded.

Each current package with its version, size and SHA-256, and the earlier builds for the record.

PackagePlatformVersionReleasedSizeSHA-256Notes
Qwen3.5 · Qwen3.6NVIDIA1.1.42026-10-3156.7 MB0d925ad92f450e833367834924e4d5d078edab7c546499e214f17b476a930f5eRelease notes
Stable Diffusion 3.5NVIDIA1.0.12026-11-0835.7 MB81a6629d8aa26cde50b2c934509be28b11df11f890e50a88528572a27141fcdeRelease notes
Wan 2.1 · Wan 2.2NVIDIA1.2.932026-11-147.2 MBc35b003d8a0a493b5b79904dd28bd1b793293061383fe6de3c498435dc5c6087Release notes
FLUX.1NVIDIA1.0.52026-11-0845.3 MB33921b9c0d6690fff86f53be87583cc1c02ad5ddacfa4af4b2e06b350da1a62aRelease notes
Veizik for Apple SiliconApple Silicon2026.10.02-r102026-10-106.1 MBd74d9f2f6b167679df1294325c12921709363f3d6aa75633f2edfd043ca9f678Release notes

How a download is checked

On NVIDIA, veizik install checks each package against the SHA-256 published for its release and refuses any other file.

On Apple Silicon the disk image is signed and notarised; compare its SHA-256 with the table, and shasum -a 256 -c SHA256SUMS checks its contents after you open it.

verify by hand
# Linux
$ sha256sum <file>
# macOS
$ shasum -a 256 <file>
# the digest must equal the SHA-256 in the table above
If a download fails or does not match.

A download that returns an error (such as 404) or whose SHA-256 differs is not a valid package. Do not install it. Download it again from its release page; if the digest still differs, write to support@veizik.com with the file name and the digest you got.

Previous releases

NVIDIA · 5 previous releases
VersionReleasedStatusNotes
cuda-wan-1.2.922026-10-31Supersededveizik WAN (CUDA) v1.2.92 — Wan 2.1 & Wan 2.2 video
cuda-wan-1.2.912026-10-30Withdrawnveizik WAN (CUDA) v1.2.91 — Wan 2.1 & Wan 2.2 video
cuda-wan-1.2.82026-10-28Supersededveizik WAN (CUDA) v1.2.8 — Wan 2.1 & Wan 2.2 video
cuda-qwen-1.1.32026-10-28Withdrawnveizik Qwen (CUDA) v1.1.3 — Qwen3.5 / Qwen3.6, single-file .vzpack
cuda-qwen-1.1.22026-10-27Withdrawnveizik Qwen (CUDA) v1.1.2 — Qwen3.5 / Qwen3.6, multimodal

Previous builds are listed for the record. A withdrawn build is not offered for download.

Apple Silicon · 11 previous releases
VersionReleasedStatusNotes
2026.10.02-r6.52026-10-10WithdrawnVeizik for Apple Silicon
2026.10.02-r6.42026-10-10WithdrawnVeizik for Apple Silicon
2026.10.02-r6.32026-10-09SupersededVeizik for Apple Silicon
2026.10.02-r6.22026-10-09WithdrawnVeizik for Apple Silicon
2026.10.02-r6.12026-10-05WithdrawnVeizik for Apple Silicon
2026.10.02-r62026-10-05WithdrawnVeizik for Apple Silicon
2026.10.02-r52026-10-04WithdrawnVeizik for Apple Silicon
2026.10.02-r32026-10-02WithdrawnVeizik for Apple Silicon
2026.10.02-r22026-10-02SupersededVeizik for Apple Silicon
2026.10.022026-10-02SupersededVeizik for Apple Silicon

Showing the 10 most recent of 11. All releases

Previous builds are listed for the record. A withdrawn build is not offered for download.

Quickstart All releases Supported models