GGUF builds for llama.cpp, Ollama, LM Studio. Mac, Windows, Linux. Multiple quant levels.
Outlier
company
AI & ML interests
None defined yet.
Recent Activity
View all activity
Organization Card
Outlier
Local AI on your Mac. No API token caps from Outlier. Free Nano + Lite, $20/mo Pro gets you the full stack, $200 once for lifetime Founders.
outlier.host · Download Mac app
Shipping tiers (v1.8)
| Tier | Plan | Disk | Min RAM |
|---|---|---|---|
| Outlier Nano (Qwen3.5-4B, MLX 4-bit) | Free | 2.37 GB | 6 GB |
| Outlier Lite (Qwen3.5-9B, MLX 4-bit) | Free | 5.04 GB | 12 GB |
| Outlier Quick (Gemma-4-26B-A4B) | Pro | 15.61 GB | 16 GB |
| Outlier Core (Qwen3.6-27B) | Pro | 15.13 GB | 24 GB |
| Outlier Code (Core weights, code config) | Pro | 15.13 GB | 24 GB |
| Outlier Plus (Qwen3.5-397B-A17B, paged) | Pro | 209 GB | 32 GB (64 GB recommended) |
| Outlier Vision (Qwen3.6-35B-A3B) | Pro | 19.0 GB | 24 GB |
Verified benchmarks live on outlier.host (with source files, exact commands, sample size, and standard error per measurement).
Patents
Three U.S. provisional applications filed April 2026, covering ternary Mixture-of-Experts inference, paged expert routing on consumer hardware, and zero-delta initialization with residual error correction. Total filing cost $195. 12-month conversion clock per application.
Research artifacts
The V3.2 family (10B / 40B / 70B / 150B) on this org are research artifacts from the training pipeline that shipped in v1.8. They are not the shipping product — the shipping product is the Mac app at outlier.host. The earlier Outlier-10B, Outlier-10B-V2, and Outlier-40B repos are preserved for continuity and download-history value.
Built by
Matt Kerr · Kerr & Company LLC · Grand Rapids, MI · matt@outlier.host
Six repos worth trying first: research flagship (70B V3.3), MLX shipping, Qwen3-Coder canonical, GGUF flagship, and V3.3 research.
-
Outlier-Ai/Outlier-70B-V3.3
Text Generation • Updated • 1.39k • 1 -
Outlier-Ai/Outlier-Lite-7B-MLX-4bit
Text Generation • 1B • Updated • 123 -
Outlier-Ai/Qwen3-Coder-30B-A3B-Instruct-MLX-4bit
Text Generation • 31B • Updated • 1.76k -
Outlier-Ai/Outlier-Max-32B-GGUF
Text Generation • 33B • Updated • 465 • 1
GGUF builds for llama.cpp, Ollama, LM Studio. Mac, Windows, Linux. Multiple quant levels.
Six repos worth trying first: research flagship (70B V3.3), MLX shipping, Qwen3-Coder canonical, GGUF flagship, and V3.3 research.
-
Outlier-Ai/Outlier-70B-V3.3
Text Generation • Updated • 1.39k • 1 -
Outlier-Ai/Outlier-Lite-7B-MLX-4bit
Text Generation • 1B • Updated • 123 -
Outlier-Ai/Qwen3-Coder-30B-A3B-Instruct-MLX-4bit
Text Generation • 31B • Updated • 1.76k -
Outlier-Ai/Outlier-Max-32B-GGUF
Text Generation • 33B • Updated • 465 • 1
models 42
Outlier-Ai/outlier-flagship-qwen36-35b-a3b-v1
Text Generation • 35B • Updated • 33
Outlier-Ai/Outlier-Compact-27B-MLX-4bit
Text Generation • 4B • Updated • 50
Outlier-Ai/Outlier-Quick-26B-MLX-4bit
Text Generation • 5B • Updated • 68
Outlier-Ai/Outlier-Core-27B-MLX-4bit
Text Generation • 4B • Updated • 74
Outlier-Ai/Outlier-Lite-9B-MLX-4bit
Text Generation • 1B • Updated • 75
Outlier-Ai/Qwen3-4B-MLX-4bit
Text Generation • 0.6B • Updated • 80
Outlier-Ai/Outlier-Nano-4B-MLX-4bit
Text Generation • Updated • 81
Outlier-Ai/Outlier-Vision
6B • Updated • 84
Outlier-Ai/Outlier-Code-27B-MLX-4bit
Text Generation • 4B • Updated • 90
Outlier-Ai/gemma-3-4b-it-MLX-4bit
Text Generation • 0.6B • Updated • 128
datasets 0
None public yet