Timon Harz

local ai. building @pocketllm

[Timon Harz @tiharz]

Congrats to all #SwiftStudentChallenge winners and applicants! It's such a privilege to have the opportunity to experience dub dub live as a 15 year old. My app playground is an app for deafblind individuals. The app translate text to morse code to haptic feedback.

[Beer @BeerBehavior]

holy no return offer. no belt, pants too long, and nikes lol #nobid #geed

[Timon Harz @tiharz]

@sammgrowth stack

[Timon Harz @tiharz]

@Parmarsg @Alexintosh @anemll can you add Gemma 4 27b by any chance?

[Timon Harz @tiharz]

@JadDTweets congrats! are you running meta ads?

[Timon Harz @tiharz]

@radiofun8 code?

[Timon Harz @tiharz]

@kevyyarthedev Update will likely come out tomorrow. Apple has long review waiting queues right now.

[Timon Harz @tiharz]

Demo of Gemma 4 4b running on an iPhone 14 using MLX with TurboQuant.

Gemma 4 beats model 4 times its size and has strong visual understanding.

[Jukan @jukan05]

The fact that memory stocks are crashing because of Google’s Turboquant is a pretty good indicator of how many clueless people this market is filled with. It’s like saying Aramco should crash because Toyota came out with a next-generation hybrid engine.

[Timon Harz @tiharz]

@jeiting https://t.co/KBfxBDqJ92

[Timon Harz @tiharz]

@ThineAI Thine

[Timon Harz @tiharz]

@raphadevelops looks really good, how are you doing that?

[Timon Harz @tiharz]

Qwen 3.5 2B is almost as good as GPT4o and it runs very well on older iPhone models. This is the first glimpse of super-intelligence in your pocket!

[Awni Hannun @awnihannun]

According to benchmarks Qwen3.5 4B is as good as GPT 4o.

GPT 4o came out ~2 years ago (May 2024).

Qwen 3.5 4B runs easily on modern mobile devices.

So the gap between frontier intelligence in a datacenter and running a model of equal quality on your iPhone could be 2-3 years. (Probably closer to 3 assuming Qwen3.5 4B is more benchmaxxed than 4o)

I don't expect the trend of increasing intelligence-per-watt to change. So in 2-3 years it's plausible we will be running GPT 5.x quality models on an iPhone. Pretty wild.

[Timon Harz @tiharz]

Qwen 3.5 is a must-have for every iPhone!
Here is Qwen 3.5 0.8b running on an iPhone 14 using MLX:

[Timon Harz @tiharz]

The new Qwen 3.5 by @Alibaba_Qwen running on-device on iPhone 14. Qwen 3.5 performance is on par with gpt-5 on some benchmarks! The 2B 6-bit model we have here is running on MLX, which is specially designed for Apple Silicon.
@pocketllm

[Timon Harz @tiharz]

@johnwalters_sd @aswinrajkailath @ycombinator @Heypocket @AkshayNarisetti @gabrieldymowski but it doesnt directly encourage users to do so. the marketing of pocket mentions recording others

[Timon Harz @tiharz]

@aswinrajkailath @ycombinator @Heypocket @AkshayNarisetti @gabrieldymowski App Store guidelines. Apple won't approve an app that is meant to record other people without them knowing.

[Timon Harz @tiharz]

@franjohn21 @blakeandersonw the problem with the local model is that it is strongly moderated and safety guardrails will be triggered for the most basic prompts