● Covered by 1 source · 1 reportMedium impact1 neutral
Swiftlet enables 80B Qwen models to run on Mac with 4.3GB RAM, 35B on iPhone
Key points
- Swiftlet runs Qwen3-Next and Qwen3.5/3.6 MoE models on Apple Silicon.
- An 80B Qwen model uses 4.3 GB RAM on Mac, a 35B model uses 2.5 GB RAM on iPhone.
- The runtime streams Mixture-of-Experts weights from storage on demand.
- The 35B model runs on iPhone 17 at approximately 1 token per second.
On-Device LLM Execution
Swiftlet is a new runtime built with Swift and Metal that enables the execution of large Qwen3-Next and Qwen3.5/3.6 Mixture-of-Experts (MoE) models on Apple devices. This includes Macs with Apple Silicon and iPhones running iOS 17 or later. The runtime achieves this by keeping only the dense core of the model in memory and streaming the MoE weights from storage as needed.
Memory Efficiency and Performance
The 80B Qwen model can run on a Mac using approximately 4.3 GB of RAM, while the 35B version operates on an iPhone 17 with about 2.5 GB of RAM. The 35B model generates output at roughly 1 token per second on an iPhone. Both models produce correct and validated output, with current development focused on optimizing kernel speed.
Implications for Mobile AI
This marks the first time a model of this scale has run natively on a phone, demonstrating a significant step towards more powerful on-device AI capabilities. While the models chat and write like large models, their factual recall is limited to that of smaller models due to only about 3 billion parameters being active per token. The 35B model is available on iPhone via the Priv AI app's Experimental Models feature, which streams the model from storage for on-device chat without server interaction.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
Primary sources
GitHub cmc_internal/api GitHub github/collect GitHub leonickson1/Swiftlet GitHub _private/browser GitHub get-started/accessibility GitHub open-source/sponsorsReporting from
Swiftlet, a Swift + Metal runtime, allows 35B and 80B Qwen models to run on Apple devices, including an iPhone 17, by streaming Mixture-of-Experts weights from storage. This development makes large language models more accessible on consumer hardware, potentially expanding the use cases for on-device AI applications.