Local AI models for Mac
The models Outmake picks for Macs with Apple silicon, with what each one takes to download. Outmake downloads any of them in one click, and finds more on Hugging Face and Civitai.
A chat model needs about its download size in memory while it runs, plus room for the conversation and for the rest of your Mac. Before you download one, Outmake says whether it runs great, runs well, is a tight fit or is too big for your particular Mac. For mixture-of-experts models, "active" is the part of the model that works on each word: they need memory for all of it, and write faster than a model of the same total size.
Chat models
| Model | Parameters | Context | Download | Commercial use |
|---|---|---|---|---|
| Qwen3.5 0.8BAlibaba Qwen · Mar 2026 · Reads pictures | 0.8B | 256K | 0.7 to 0.8 GB | YesApache 2.0 |
| Qwen3.5 2BAlibaba Qwen · Mar 2026 · Reads pictures | 2B | 256K | 1.8 to 2.2 GB | YesApache 2.0 |
| Qwen3.5 4BAlibaba Qwen · Mar 2026 · Reads pictures | 4B | 256K | 3.0 to 4.2 GB | YesApache 2.0 |
| Qwen3.5 9BAlibaba Qwen · Mar 2026 · Reads pictures | 9B | 256K | 5.6 to 8.4 GB | YesApache 2.0 |
| Gemma 4 E2BGoogle · Mar 2026 · Reads pictures | 5.1B | 128K | 3.5 to 5.5 GB | YesApache 2.0 |
| Gemma 4 E4BGoogle · Mar 2026 · Reads pictures | 8B | 128K | 7.6 GB | YesApache 2.0 |
| Gemma 4 26BGoogle · Mar 2026 · Reads pictures | 26B, 3.8B active | 256K | 14 to 28 GB | YesApache 2.0 |
| Gemma 4 31BGoogle · Mar 2026 · Reads pictures | 31B | 256K | 16 to 26 GB | YesApache 2.0 |
| Qwen3.8 27BAlibaba Qwen · Aug 2026 · Reads pictures | 28B | 256K | 15 to 23 GB | YesApache 2.0 |
| Devstral Small 2 24BMistral AI · Nov 2025 · Reads pictures | 24B | 256K | 12 to 20 GB | YesApache 2.0 |
| GLM 4.7 FlashZ.ai · Jan 2026 | 31B, 3B active | 198K | 16 to 25 GB | YesMIT |
| GPT-OSS 20BOpenAI · Aug 2025 | 21B, 3.6B active | 128K | 12 to 13 GB | YesApache 2.0 |
| GPT-OSS 120BOpenAI · Aug 2025 | 117B, 5.1B active | 128K | 63 to 64 GB | YesApache 2.0 |
| Qwen3.6 35BAlibaba Qwen · Apr 2026 · Reads pictures | 36B, 3B active | 256K | 19 to 30 GB | YesApache 2.0 |
| Qwen3.5 397BAlibaba Qwen · Feb 2026 · Reads pictures | 403B, 17B active | 256K | 124 to 246 GB | YesApache 2.0 |
| Qwen3.5 122BAlibaba Qwen · Mar 2026 · Reads pictures | 122B, 10B active | 256K | 58 to 113 GB | YesApache 2.0 |
| Qwen3.8 Flash NextAlibaba Qwen · Sep 2026 · Reads pictures | 177B, 6B active | 256K | 91 to 170 GB | With conditionsQwen Community License 1.0 |
| DeepSeek V4 FlashDeepSeek · Jul 2026 | 284B, 13B active | 1M | 128 to 162 GB | YesMIT |
| MiniMax M2.7MiniMax · Apr 2026 | 229B, 10B active | 192K | 102 to 169 GB | NoMiniMax M2.7 Non-Commercial License |
| GLM 5.3Z.AI · Aug 2026 | 754B, 40B active | 1M | 254 to 343 GB | With conditionsGLM-5.3 License |
| Kimi K2.6Moonshot AI · Apr 2026 · Reads pictures | 1T, 32B active | 256K | 341 GB | With conditionsModified MIT |
Picture models
| Model | Parameters | Download | Commercial use |
|---|---|---|---|
| Z-Image TurboTongyi (Alibaba) · Nov 2025 | 6B | 6.5 to 11 GB | YesApache 2.0 |
| FLUX.1 SchnellBlack Forest Labs · Aug 2024 | 12B | 17 to 23 GB | YesApache 2.0 |
| Qwen ImageAlibaba Qwen · Aug 2025 | 20B | 18 to 30 GB | YesApache 2.0 |
| Qwen Image EditAlibaba Qwen · Sep 2025 | 20B | 19 to 31 GB | YesApache 2.0 |
| FLUX.2 Klein 4BBlack Forest Labs · Dec 2025 | 4B | 5.3 to 8.9 GB | YesApache 2.0 |
Video models
| Model | Parameters | Download | Commercial use |
|---|---|---|---|
| Wan 2.2 5BWan (Alibaba) · Jul 2025 | 5B | 9.5 to 11 GB | YesApache 2.0 |
| Wan 2.2 14B Image to VideoWan (Alibaba) · Jul 2025 | 14B | 21 to 35 GB | YesApache 2.0 |
| Wan 2.2 14B Text to VideoWan (Alibaba) · Jul 2025 | 14B | 23 to 36 GB | YesApache 2.0 |
| Wan 2.1 14B Text to VideoWan (Alibaba) · Feb 2025 | 14B | 14 to 20 GB | YesApache 2.0 |
| LTX-2.3Lightricks · Mar 2026 | 22B | 24 to 34 GB | With conditionsLTX-2 Community License |
| Wan 2.1 1.3BWan (Alibaba) · Feb 2025 | 1.3B | 6.7 GB | YesApache 2.0 |
Licence conditions
- Qwen3.8 Flash Next
- Commercial use is allowed, but a product or service with more than 100 million monthly active users or $20 million in monthly revenue must display the model name on its interface, and running it as a hosted Model-as-a-Service or AI Work Assistant business needs a separate license from Qwen.
- GLM 5.3
- Custom MIT-style license from Z.AI. Commercial use is allowed; only a Model-as-a-Service business with more than $10 billion in aggregate revenue over any 12 months needs to pass a Z.AI security review before commercial use.
- Kimi K2.6
- MIT-style license. A commercial product or service with more than 100 million monthly active users or $20 million in monthly revenue must display "Kimi K2.6" on its interface.
- LTX-2.3
- Free to use, for work too, for people and companies with under $10 million in yearly revenue. A bigger company needs a paid license from Lightricks. Its use policy bans, among other things, impersonating real people, false information and weapons work.
Download sizes add up every file a model needs, at the smallest and largest quantization Outmake offers for it. A licence's summary can't carry every condition, so read the full licence on the model's page before commercial work. This list is Outmake's catalog as of October 6, 2026. In the app the catalog updates every day, and Search finds other models on Hugging Face and Civitai.
How to run Civitai models on a Mac · Models on this Mac, in Help