On this page
You can run DeepSeek on your iPhone with Off Grid AI Mobile (OGAM). It’s a small distilled DeepSeek R1 model, about 1 GB on any recent iPhone or 5 GB on an 8 GB one, and it answers in airplane mode with nothing uploaded.
The official DeepSeek app can’t do that, because it sends every message to DeepSeek’s servers. And the full DeepSeek V4.1 Flash doesn’t fit on any phone. If you want that one, skip to the full model.
App Store · Download Off Grid AI
Which DeepSeek fits on an iPhone?
DeepSeek’s newest model, DeepSeek V4.1 Flash, has 552B parameters. It only uses 8B to 16B of them per token, but every one still has to sit in memory. The official weights are about 510 GB, and the smallest community GGUF (unsloth’s UD-IQ1_S) is still about 233 GB. The biggest iPhone has 12 GB.
DeepSeek hasn’t released a small V4 or V4.1 model. The small DeepSeek models you can run are the R1 distills, built by teaching Qwen and Llama models to reason like DeepSeek R1. They’re MIT licensed and come straight from DeepSeek’s Hugging Face page.
| Model | Built on | Q4_K_M file | iPhone RAM |
|---|---|---|---|
| DeepSeek-R1-Distill-Qwen-1.5B | Qwen2.5-Math-1.5B | 1.12 GB | 6 GB is fine |
| DeepSeek-R1-Distill-Qwen-7B | Qwen2.5-Math-7B | 4.68 GB | 8 GB (Tight), 12 GB runs well |
| DeepSeek-R1-0528-Qwen3-8B | Qwen3-8B | 5.03 GB | 8 GB (Tight), 12 GB runs well |
| DeepSeek V4.1 Flash | its own 552B design | about 233 GB at the smallest | doesn’t fit |
File sizes are from unsloth’s GGUF builds on Hugging Face.
Pick the 8B on a 12 GB iPhone, like the iPhone 17 Pro Max. DeepSeek-R1-0528-Qwen3-8B is the newest small model DeepSeek has put out, and on DeepSeek’s own model card it beats the plain Qwen3-8B at maths (86.0 against 76.0 on AIME 2024). On an 8 GB iPhone it still loads, but it leaves little room, so close other apps first. The 1.5B is the safe pick for 6 GB iPhones. It’s built on a maths model, so it’s good at working through a problem step by step and weaker at general chat.
How Off Grid AI decides what your iPhone can run
You don’t have to do the maths. OGAM’s model search hides anything too big for your iPhone and marks files that only just fit as Tight. On iPhone, a file up to 60% of the RAM iOS reports runs well, and one up to 70% is Tight, because iOS maps the model from storage instead of copying it all into the app’s memory. Past 70%, it isn’t offered.
iOS reports a little less than the number on the box. A 12 GB iPhone 17 Pro Max reports 11.2 GB.
Set it up
- Install OGAM from the App Store. You need iOS 17 or later. It’s free, with no account.
- Tap Models in the tab bar. Stay on the Text tab.
- Type deepseek r1 in Search models.
- Tap DeepSeek-R1-0528-Qwen3-8B, or DeepSeek-R1-Distill-Qwen-1.5B on a 6 GB iPhone.
- Under Sizes, take Recommended. It’s the best file for your iPhone. Download it on Wi-Fi.
- When it finishes, tap Chat now.
Turn on airplane mode and send another message. It still answers, because the model runs on your iPhone’s GPU through Metal.
R1 models think before they answer, so you’ll see a reasoning step first. Give it a problem with a right answer, like “I earn 4,200 a month, rent is 1,450 and bills are 380. How much is left after 12 months?”, and check the working.
What about the full DeepSeek V4.1 Flash?
It can’t run on your iPhone, but your iPhone can talk to a server that runs it. There are two honest ways.
DeepSeek’s own app or API. It’s cheap and it’s the real model. DeepSeek’s API serves V4.1 Flash as deepseek-flash, from $0.30 per million input tokens at peak and half that off-peak. But DeepSeek’s privacy policy says it stores your data in China and may use what you type to train its models. You can opt out of training by emailing them.
OGAM can use the API so you keep one app. Go to Settings, then Remote servers, then Add manually. Put https://api.deepseek.com/v1 in Address, paste your DeepSeek key in API key and type deepseek-flash in Text model. Tap Test connection, then Add server. The key stays in the iPhone’s Keychain. The app warns you that this server is on the public internet, and it’s right: those messages go to DeepSeek.
Off Grid AI Boost. Boost runs open models, DeepSeek V4.1 Flash included, on our own GPUs for one flat price from $4.99 a month for the first cohort. We don’t keep your prompts or train on them. It’s a pre-order today: API keys go out by January 2027, and usage limits for each plan aren’t published yet.
If something goes wrong
| Symptom | Check | Fix |
|---|---|---|
| The 8B doesn’t show in search | Your iPhone’s RAM | It’s too big for this iPhone. Take the 1.5B. |
| The model shows Tight | Other apps holding memory | It loads, but close heavy apps first or pick Smaller. |
| The model won’t load | Free memory right now | Your current model stays loaded. Close apps and try again. |
| The answer rambles for ages | It’s a reasoning model | Ask a narrower question. The 1.5B is weakest at open chat. |
| The DeepSeek server test fails | The address and key | Use https://api.deepseek.com/v1 and a key with credit on it. |
FAQ
Can I run DeepSeek V4.1 Flash on my iPhone? No. DeepSeek V4.1 Flash has 552B parameters and its smallest community GGUF build is about 233 GB. An iPhone has 6 to 12 GB of RAM. What runs on an iPhone is a small distilled DeepSeek R1 model, such as DeepSeek-R1-Distill-Qwen-1.5B or DeepSeek-R1-0528-Qwen3-8B.
Which DeepSeek model fits my iPhone? DeepSeek-R1-Distill-Qwen-1.5B at Q4_K_M, a 1.12 GB file, fits any iPhone with 6 GB of RAM. DeepSeek-R1-0528-Qwen3-8B at Q4_K_M, a 5.03 GB file, runs well on a 12 GB iPhone and shows as Tight on an 8 GB one. Off Grid AI hides any model too big for your iPhone.
Does the DeepSeek app work offline on iPhone? No. The official DeepSeek app sends every message to DeepSeek’s servers. To use DeepSeek offline, download a distilled model into Off Grid AI. It runs on the iPhone itself, so it keeps working in airplane mode.
Is it free? Yes. Off Grid AI is free on the App Store, with no account and no message limit. The DeepSeek model files are MIT licensed and free to download from Hugging Face.
How do I use the full DeepSeek V4.1 Flash from my iPhone? Through a server. DeepSeek’s own app and API run it on DeepSeek’s servers, which store data in China and may use it for training. Off Grid AI Boost will run DeepSeek V4.1 Flash on our GPUs at a flat price and doesn’t keep your prompts; it’s a pre-order, with API keys going out by January 2027.
Related guides
- How to Find AI Models Your iPhone Can Actually Run in 2026 (No Failed Downloads)
- How to Run Qwen 3.5 on Your iPhone in 2026 (Completely Offline, No Subscription)
- How to Run DeepSeek on Your Android Phone in 2026 (Offline, Free, No Account)
- A Flat-Price, Private API for Open Models in 2026
Get Off Grid AI on the App Store and run DeepSeek on your own iPhone today.