---

VaultAI isn't one AI — it gives you 21 to 46 of them depending on the tier, plus image generation, all working from a single superfast SSD drive that runs completely offline. You never have to memorize any of this: you can let VaultAI automatically pick the right model for your task. But if you're wondering how each model answers differently, we gave them all the following prompt: “Owning your AI is like ___.” You can see below how each model answered it in its own style and capability.

VaultAI Chat & Reasoning Models

The all-rounders you'll talk to most. They run from featherweight to flagship; VaultAI reaches for the right one automatically.

Qwen3-235B-A22B 235B

The giant of the drive. With 235 billion parameters it holds the deepest knowledge and sharpest reasoning of anything here — yet a clever design fires only a slice at a time, so it stays usable on a high-memory machine. Reach for it when the stakes are high.

GPT-OSS 120B 120B

A 120-billion-parameter powerhouse built for serious thinking — your hardest reasoning, deepest research, most demanding code, all kept on your drive.

DeepSeek-R1 70B Uncensored 70B

Full-strength reasoning with the guardrails off — a model that genuinely thinks out loud and never refuses or lectures. Seventy billion parameters answering to one person: you.

IBM Granite 4 Small-H 32B 32B · 1M context

Its superpower is memory — roughly a million words at once, so you can feed it an entire codebase or a whole book and ask across all of it.

Qwen3.6 35B-A3B 35B

A do-everything workhorse that feels far quicker than its size suggests. It codes, reasons, reads images, uses tools, and remembers a quarter-million words.

Qwen3 32B Uncensored 32B

Big-model reasoning across dozens of languages, with nothing off the table.

DeepSeek-R1 32B 32B

Cloud-grade logic at a speed you'll actually enjoy — careful, show-your-work reasoning, quick enough for everyday use.

GLM-4 32B 32B

An open 32B that reasons and writes at the level of much larger models. Fully open-licensed and entirely private.

Mistral Small 3.2 24B 24B

A polished, dependable all-rounder that shines at following instructions and connecting to tools. It reads images, too.

GPT-OSS 20B 20B

The everyday champion — fast, capable, and happy to handle the bulk of what you throw at it.

GPT-OSS 20B Abliterated 20B

The same fast, capable everyday model, with the refusals stripped out.

Gemma 4 26B MoE 26B

Google's design lets this 26B model run at the speed of a 4B one — bigger answers without the wait, plus a quarter-million words of context.

GLM-4 9B 9B

A compact, capable all-rounder — quick, sharp general chat in a small footprint. Fully open-licensed and private to your drive.

Gemma 4 E4B 8B

A higher-quality take on the default assistant — light enough to run almost anywhere, with image understanding and tool use built in.

Ministral 3 8B 8B

A laptop-friendly all-rounder that sees images and remembers a quarter-million words — rare reach for something this small and quick.

Hermes 3 8B Uncensored 8B

A nimble, uncensored model with a real flair for creative writing.

Dolphin 3 8B Uncensored 8B

Widely considered one of the best small uncensored models there is — fast, capable, refusal-free.

Qwen3.5 9B 9B

A capable mid-weight that's especially tidy when you need structure — clean tables, formatted documents — while still handling heavier reasoning.

Qwen3.5 9B Uncensored 9B

The same capable mid-weight with restrictions lifted — retuned for stable, refusal-free long-form answers.

Mistral 7B 7B

The dependable utility model — quick, even-tempered, good at doing exactly what you ask.

Qwen3 4B 4B

A small, speedy model that's surprisingly good at using tools — a nimble choice for lighter machines.

Ministral 3 3B 3B

A featherweight that still sees images and remembers a quarter-million words — remarkable reach for something this small and fast.

IBM Granite 4 Tiny-H 7B 7B · 1M context

A small model with an enormous memory — up to a million words at once — that runs comfortably on modest hardware.

IBM Granite 4 Micro 3B 3B

An efficient little tool-caller built to run anywhere.

Gemma 4 E2B 5.1B

The featherweight all-rounder — fast on almost any computer, sees images, tuned for smooth everyday conversation.

VaultAI Coding Models

Built to read, write, and debug code — privately, with nothing about your projects ever leaving the drive.

Ornith 1.0 35B 35B

The most capable coder on the drive — it plans and works through multi-step coding jobs like a senior engineer, rivaling the big cloud models, all private.

Qwen3-Coder 30B 30B

The coding flagship — it carries a whole task across a large project, remembering a quarter-million words of context.

Codestral 22B 22B

A specialist that's almost spooky at finishing your code — over 95% on the standard fill-in-the-blank test.

Ornith 1.0 9B 9B

Punches three sizes above its weight — coding skill you'd expect from a far larger model, small and fast enough to run anywhere.

Devstral Small Lite light

A lightweight helper for quick completions and small fixes — fast, low-overhead, always there.

VaultAI Vision Models

Models that can actually see. Hand them a photo, a screenshot, a scanned page, or a chart and ask away.

Qwen3-VL 32B 32B

The serious one for visual work — it studies a complex image and reasons about what's in it, not just describes it.

Qwen3-VL 8B 8B

Sharp visual understanding at a lighter weight — a strong everyday pick for questions about images and screenshots.

MiniCPM-V 8B 8B

The reader of the group — especially good at pulling text out of images and making sense of documents.

LLaVA 7B 7B

A reliable, well-rounded model for talking about images.

Moondream 1.8B 1.8B

Featherweight vision for quick looks — a fast answer about an image without spinning up anything heavy.

Gemma 3 4B 4B

A quick, light model that can also look at pictures — great for fast questions and a snap read of a photo on modest hardware.

VaultAI Medical Models

Purpose-built medical models for private health questions that never touch a server. Great for understanding — not a substitute for your doctor.

MedGemma 27B 27B

The most capable medical model on the drive, built for clinical-level reasoning and biomedical analysis — kept entirely private.

MedGemma 1.5 4B 4B

A compact medical model that can also read images — X-rays, skin photos, pathology slides — and answer health questions.

MedGemma 4B Lite 4B

The same medical focus, trimmed to run on minimal hardware — quick, private health questions on any machine.

VaultAI Specialty & Uncensored Models

A handful of models for specific jobs and unrestricted work.

MN-Darkest 29B 29B

A storyteller — tuned for fiction and roleplay, a creative collaborator that won't flinch.

Qwen2.5 14B Uncensored 14B

An unrestricted model for research and creative work.

Claude-Gemma3 12B 12B

Tuned to answer in the style of Claude — helpful, thorough, carefully explained.

GPT-NeoX 20B Code 20B

An open model geared toward generating and analyzing code — a useful alternative voice for programming.

Xortron Criminal specialist

A niche tool for writers, tuned for crime fiction, mystery, and forensic storytelling.

Zephyr 7B Lite 7B

An exceptionally light, helpful chat model that barely sips resources.

VaultAI Image Generation Engines

It's not just words. VaultAI makes pictures too, right on the drive, with nothing sent to the cloud.

Z-Image-Turbo 6B · every tier

A fast, photorealistic image generator — describe what you want and it renders it in seconds, created locally, nothing uploaded.

FLUX.2 Klein 4B 4B · Pro and Ultra

A nimble image engine that produces high-quality results in just a few steps, with a lighter footprint.

---

That's up to 46 AI models plus image generation — chat, reasoning, code, vision, medical, and creative — delivered on a single superfast drive. No accounts, no subscriptions, no internet required, and nothing you do ever leaves your hands.