Local AI / Model Catalog & Launcher

Find a model
that fits.

Start with the work you want to do. Compare models and memory needs, then copy a command to run the model on your own machine.

Your path to local AI

  1. Choose a modelDocuments, code, chat, vision or video
  2. Check memoryCompare a planning estimate with your hardware
  3. Run it locallyGet weights or setup instructions from the upstream source

02 / Ollama builds

Find a build for your hardware.

Install Ollama to use the commands below. Builds are sorted by the original repository’s 30-day Hugging Face downloads. Sizing examples show where more memory changes your options.

Curated model builds

How we estimate memory

Alibaba Cloud

General

Qwen3 14B

A multilingual reasoning model for writing, conversation and structured responses.

  • 14.8B
  • GGUF · Q4_K_M
  • Apache 2.0
Download size
9.3 GB

13,025,679 Hugging Face downloads · 30 days

Run with Ollama

ollama run qwen3:14b-q4_K_M
Download only
ollama pull qwen3:14b-q4_K_M

Alibaba Cloud

Vision

Qwen3.8 27B

Released

A text and vision model for coding, document review and image understanding.

  • 27B
  • GGUF · Q4_K_M
  • Apache 2.0
Download size
17.7 GB

6,757,964 Hugging Face downloads · 30 days

Run with Ollama

ollama run qwen3.8:27b-q4_K_M
Download only
ollama pull qwen3.8:27b-q4_K_M

Google DeepMind

Vision

Gemma 4 31B

A multimodal model for screenshots, visual documents and image reasoning.

  • 31B
  • GGUF · Q4_K_M
  • Apache 2.0
Download size
20 GB

6,520,176 Hugging Face downloads · 30 days

Run with Ollama

ollama run gemma4:31b-it-q4_K_M
Download only
ollama pull gemma4:31b-it-q4_K_M

Alibaba Cloud

Documents

Qwen3 8B

A compact multilingual text model for summarization and answering questions from document excerpts.

  • 8.2B
  • GGUF · Q4_K_M
  • Apache 2.0
Download size
5.2 GB

6,111,761 Hugging Face downloads · 30 days

ZXO Pilot · 8 of 8 document Q&A checks passed on a Windows test host

Run with Ollama

ollama run qwen3:8b-q4_K_M
Download only
ollama pull qwen3:8b-q4_K_M

OpenAI

General

gpt-oss-20b

A smaller reasoning model for local text workflows and structured responses.

  • 20.9B total · 3.6B active
  • GGUF · MXFP4
  • Apache 2.0
Download size
14 GB

5,641,053 Hugging Face downloads · 30 days

Run with Ollama

ollama run gpt-oss:20b
Download only
ollama pull gpt-oss:20b

OpenAI

General

gpt-oss-120b

A larger reasoning model for local assistants, tool use and text workflows.

  • 117B total · 5.1B active
  • GGUF · MXFP4
  • Apache 2.0
Download size
65 GB

2,431,461 Hugging Face downloads · 30 days

Run with Ollama

ollama run gpt-oss:120b
Download only
ollama pull gpt-oss:120b

Alibaba Cloud

Coding

Qwen2.5-Coder 14B

A model for generating, explaining and revising code with developer review.

  • 14.7B
  • GGUF · Q4_K_M
  • Apache 2.0
Download size
9 GB

1,795,901 Hugging Face downloads · 30 days

Run with Ollama

ollama run qwen2.5-coder:14b-instruct-q4_K_M
Download only
ollama pull qwen2.5-coder:14b-instruct-q4_K_M

DeepSeek

General

DeepSeek R1 671B

The full mixture-of-experts model, included as a large memory example for multi-GPU planning.

  • 671B total · 37B active
  • GGUF · Q4_K_M
  • MIT
  • Sizing example
Download size
404 GB

1,646,817 Hugging Face downloads · 30 days

Run with Ollama

ollama run deepseek-r1:671b-q4_K_M
Download only
ollama pull deepseek-r1:671b-q4_K_M

Alibaba Cloud

Coding

Qwen2.5-Coder 32B Instruct

A larger coding model for testing code generation, explanation and review workflows.

  • 32.8B
  • GGUF · Q4_K_M
  • Apache 2.0
  • Sizing example
Download size
20 GB

542,448 Hugging Face downloads · 30 days

Run with Ollama

ollama run qwen2.5-coder:32b-instruct-q4_K_M
Download only
ollama pull qwen2.5-coder:32b-instruct-q4_K_M

Ornith AI

General

Ornith 1.5 122B

Released

A large mixture-of-experts model for reasoning and comparing shared-memory and multi-GPU systems.

  • 122B
  • GGUF · Q4_K_M
  • MIT
  • Sizing example
Download size
76.8 GB

328,590 Hugging Face downloads · 30 days

Run with Ollama

ollama run ornith-1.5:122b-q4_K_M
Download only
ollama pull ornith-1.5:122b-q4_K_M

Ricursive Intelligence

Coding

Ripple-65B Preview

Released

A dense reasoning model for evaluating code and multi-step development tasks.

  • 65B
  • GGUF · Q4_K_M
  • Apache 2.0
Download size
38.8 GB

10,328 Hugging Face downloads · 30 days

Run with Ollama

ollama run ripple-preview:65b-q4_K_M
Download only
ollama pull ripple-preview:65b-q4_K_M

03 / vLLM and SGLang

Server models

Check the deployment guide for the GPU count, runtime version and context length your project needs.

Thinking Machines Lab

Inkling 2 Small

Released

A text and image reasoning model with a long context window.

  • 594B total · 17B active
  • Safetensors · NVFP4 / MXFP8
  • Apache 2.0
Approx. weight download
394 GB

3,647 Hugging Face downloads · 30 days

Publisher sizing examples are 4 × H200 or 2 × B300. Verified SGLang recipes use 8 × H200 or 4 × B300. Confirm the runtime and GPU layout for your project.

Z.ai

GLM-5.3-Flash

Released

An FP8 model for evaluating text and image workloads on a multi-GPU server.

  • 320B total · 18B active
  • Safetensors · FP8
  • MIT
Approx. weight download
328 GB

6,382,979 Hugging Face downloads · 30 days

vLLM tested this FP8 build with text-only KV cache offloading on 4 × H200 at a 16K context limit. Confirm runtime and interconnect for a ZXO S4.

Moonshot AI

Kimi K3.1 Pro

The newer K3 family checkpoint for evaluating large multimodal workloads.

  • 2.8T total · 104B active
  • Safetensors · MXFP4 weights
  • Kimi K3.1-Pro License
Approx. weight download
1.56 TB

43,687 Hugging Face downloads · 30 days

vLLM provides an 8 × B300 recipe. Review Moonshot’s commercial terms and plan the cluster for your workload.

04 / Sizing and sources

A starting point for testing.

How is memory estimated?

We convert the registry download size from GB to GiB, add 2 GiB per 4,096 context tokens, then add the larger of 8 GiB or 20% of the model size for runtime overhead. This is a screening rule for one active request, not a model-specific KV cache calculation. Image inputs, parallel requests and some runtimes can exceed it. The estimate covers the listed Ollama build only.

How does the hardware filter work?

It shows builds that fit within our planning budget: 96 GiB for a 128 GB GB10 system, 80 GiB on a single 96 GB GPU, or 120 GiB on a single H200 NVL GPU. The shared-memory budget leaves room for the host. The W1/W2, W3 and S4 profiles represent ZXO configuration targets. W3 and S4 also show builds above the one-GPU budget as candidates for multi-GPU review when their estimate is below the combined planning allowance. GPU memory is separate on each card. Actual fit depends on the runtime, interconnect and how it splits the model.

What does ZXO measure?

A performance benchmark records the exact model build, host, context length and measured generation speed. The Qwen3 8B pilot example shows document Q&A checks. Plan to measure speed and first-token latency on the intended host.

What do download counts measure?

Counts are snapshots of the original publishers’ Hugging Face model repositories over the previous 30 days, checked . The separate H3 count is for ComfyUI’s workflow repository. Hugging Face counts requests to selected files, so these figures do not measure unique users, completed weight downloads or pulls of the listed Ollama build.

Where do weights and credentials go?

Commands run on your machine and fetch the model from the listed registry. Review each publisher’s license and terms before deployment. For a gated Hugging Face checkpoint, accept the publisher’s terms there and authenticate locally with hf auth login. ZXO does not collect access tokens or host model weights in this catalog.

05 / Deployment tools

Start with a local toolkit.

Get a sample Ollama Modelfile, vLLM and Open WebUI setup, and a benchmark script for first-token latency and generation speed. Review the prerequisites for your runtime and host.

Download the starter kit

Configuration templates. Includes no model weights or benchmark results.

06 / Local AI Pilot

Test it with your workload.

We’ll scope a four-week pilot around one workflow and agree on what to measure. Need an offline install? Tell us which models and network restrictions apply so we can review what can be packaged.

Request a Local AI Pilot See the pilot schedule and deliverables