--- thumbnail: https://huggingface.co/Outlier-Ai/README/resolve/main/assets/thumbnail.png ---

Outlier

Outlier

Ternary MoE + Apple Silicon quantization for on-device AI.
Desktop app for Mac. No token caps. Free forever.

outlier.host · Discord · Founders ($200 lifetime)

--- ## Try it now | | | |---|---| | **Download for Mac** | [outlier.host](https://outlier.host) — 9.2 MB DMG, v1.4 shipping | | **Join Discord** | [discord.gg/Hapennmdn9](https://discord.gg/Hapennmdn9) | | **Support the mission** | [Founders lifetime $200](https://buy.polar.sh/polar_cl_mJfYZsEpEMDcYrgxzvTdnahSeSQNq1UYLqV0l08CUhW) · 500-seat cap | --- ## What is Outlier? A Mac-native AI platform with curated open-weights models for Apple Silicon. Built solo in 20 days on a Mac Studio, under $1,200 of compute spend. Three U.S. provisional patents filed on the underlying ternary MoE architecture. Two tracks on HuggingFace: 1. **Research MoE** — ternary Mixture-of-Experts overlays on frozen Qwen2.5 bases. 10B, 40B, 70B, 150B scales, MMLU-verified. 2. **Apple Silicon conversions** — MLX 4-bit builds of strong open-weights models, tuned for the Outlier desktop app and usable standalone via `mlx_lm`. --- ## Research line (MMLU verified) All values at n=14,042 · lm-evaluation-harness v0.4.9.1 · bf16 5-shot · source weights unchanged since Day 13 (2026-04-13). | Scale | MMLU | Stderr | Base | Repo | |---|---|---|---|---| | Outlier-10B V3.3 | 70.87% | ±0.37% | Qwen2.5-7B-Instruct | [Outlier-Ai/Outlier-10B](https://huggingface.co/Outlier-Ai/Outlier-10B) | | Outlier-40B V3.3 | 77.80% | ±0.33% | Qwen2.5-14B-Instruct | [Outlier-Ai/Outlier-40B](https://huggingface.co/Outlier-Ai/Outlier-40B) | | Outlier-70B V3.3 (alpha-fixed) | 83.10% | ±0.30% | Qwen2.5-32B-Instruct | [Outlier-Ai/Outlier-70B-V3.3](https://huggingface.co/Outlier-Ai/Outlier-70B-V3.3) | | Outlier-150B V3.2 | 84.46% | ±0.29% | Qwen2.5-72B-Instruct | [Outlier-Ai/Outlier-150B-V3.2](https://huggingface.co/Outlier-Ai/Outlier-150B-V3.2) | **Architecture:** shared full-precision FFN plus gated ternary expert FFN per layer. Overlay checkpoints load on top of frozen Qwen2.5 bases. 70B V3.3 alpha-fix overlay is 15 KB, trained in 18 minutes on one B200, +1.61pp MMLU over V3.2. --- ## Shipping tier for Apple Silicon MLX 4-bit builds, verified on Mac Studio M1 Ultra 64GB. Bundled in the Outlier desktop app tier library. | Tier | Base | Peak RAM | Speed | Repo | |---|---|---|---|---| | Nano | Qwen3-1.7B | ~2 GB | bench pending | [Outlier-Nano-1.7B-MLX-4bit](https://huggingface.co/Outlier-Ai/Outlier-Nano-1.7B-MLX-4bit) | | Lite | Qwen2.5-7B | 4.47 GB | 71.30 tok/s | [Outlier-Lite-7B-MLX-4bit](https://huggingface.co/Outlier-Ai/Outlier-Lite-7B-MLX-4bit) | | Compact | Qwen2.5-14B | 8.24 GB | 37.26 tok/s | [Outlier-Compact-14B-MLX-4bit](https://huggingface.co/Outlier-Ai/Outlier-Compact-14B-MLX-4bit) | Plus cross-platform GGUF builds for llama.cpp / Ollama / LM Studio / Jan: [Lite 7B](https://huggingface.co/Outlier-Ai/Outlier-Lite-7B-GGUF), [Compact 14B](https://huggingface.co/Outlier-Ai/Outlier-Compact-14B-GGUF), [Max 32B](https://huggingface.co/Outlier-Ai/Outlier-Max-32B-GGUF). --- ## Apple Silicon conversions MLX 4-bit conversions of strong open-weights models, Mac-tuned, upstream-named for HF search discovery: - **DeepSeek R1 Distill** — [32B](https://huggingface.co/Outlier-Ai/DeepSeek-R1-Distill-Qwen-32B-MLX-4bit) · [14B](https://huggingface.co/Outlier-Ai/DeepSeek-R1-Distill-Qwen-14B-MLX-4bit) · [Llama-8B](https://huggingface.co/Outlier-Ai/DeepSeek-R1-Distill-Llama-8B-MLX-4bit) · [Qwen-7B](https://huggingface.co/Outlier-Ai/DeepSeek-R1-Distill-Qwen-7B-MLX-4bit) - **Qwen3** — [32B](https://huggingface.co/Outlier-Ai/Qwen3-32B-MLX-4bit) · [14B](https://huggingface.co/Outlier-Ai/Qwen3-14B-MLX-4bit) · [8B](https://huggingface.co/Outlier-Ai/Qwen3-8B-MLX-4bit) · [4B](https://huggingface.co/Outlier-Ai/Qwen3-4B-MLX-4bit) - **Qwen Coder** — [Qwen3-Coder-30B-A3B](https://huggingface.co/Outlier-Ai/Qwen3-Coder-30B-A3B-Instruct-MLX-4bit) · [Qwen2.5-Coder-32B](https://huggingface.co/Outlier-Ai/Qwen2.5-Coder-32B-Instruct-MLX-4bit) · [14B](https://huggingface.co/Outlier-Ai/Qwen2.5-Coder-14B-Instruct-MLX-4bit) · [7B](https://huggingface.co/Outlier-Ai/Qwen2.5-Coder-7B-Instruct-MLX-4bit) - **Other** — [QwQ-32B](https://huggingface.co/Outlier-Ai/QwQ-32B-MLX-4bit) · [Yi-Coder-9B](https://huggingface.co/Outlier-Ai/Yi-Coder-9B-Chat-MLX-4bit) · [Phi-4-mini](https://huggingface.co/Outlier-Ai/Phi-4-mini-instruct-MLX-4bit) · [gpt-oss-20b](https://huggingface.co/Outlier-Ai/gpt-oss-20b-MLX-4bit) · [SmolLM3-3B](https://huggingface.co/Outlier-Ai/SmolLM3-3B-MLX-4bit) · [Gemma 3 27B](https://huggingface.co/Outlier-Ai/gemma-3-27b-it-MLX-4bit) · [Gemma 3 4B](https://huggingface.co/Outlier-Ai/gemma-3-4b-it-MLX-4bit) Every conversion is a faithful port of upstream weights — capability credit belongs to the upstream authors. We add the MLX 4-bit packaging and desktop-app integration. --- ## Collections - [Outlier for Apple Silicon (MLX)](https://huggingface.co/collections/Outlier-Ai/outlier-consumer-edition-69e2fb4a0df119ea1747275e) - [Outlier Research MoE](https://huggingface.co/collections/Outlier-Ai/outlier-research-69e2fb3a71984614b3c7a279) - [Outlier Server V3.2](https://huggingface.co/collections/Outlier-Ai/outlier-server-v32-69e2fb4b71984614b3c7a4a3) --- ## Patents + citation Architecture, training pipeline, and inference engine covered by US provisional patents 64/026,886, 64/030,368, and 64/034,028 (Kerr & Company LLC, 2026). ```bibtex @misc{kerr2026outlier, title = {Outlier: Ternary Mixture-of-Experts for On-Device AI}, author = {Kerr, Matthew}, year = {2026}, url = {https://huggingface.co/Outlier-Ai} } ``` --- ## Contact Matt Kerr · [outlier.host](https://outlier.host) · [@mattkerr09](https://x.com/mattkerr09) · [matt@outlier.host](mailto:matt@outlier.host) Built solo in Grand Rapids, Michigan.