Quick answer
On September 1, 2026, Perplexity launched Hybrid Compute on Mac: a 27-billion-parameter local model (PPLX Qwen 3.8 27B, based on Alibaba's open-weight Qwen) runs on the user's machine and handles parts of each task on-device, routing the rest to Perplexity's cloud models. It is included in Perplexity Pro, Max, and Enterprise at no extra cost and uses no cloud credits for the work done locally.
This is the first time a major consumer AI product has made local-plus-cloud computing the default rather than an enthusiast option. Developers have been running models locally with Ollama and LM Studio for two years; Perplexity is doing it silently for ordinary subscribers, using the Apple silicon they already own.
How it works
- A 27B open-weight model, fine-tuned by Perplexity from Qwen 3.8, is downloaded to the Mac
- A router decides per task what the local model can handle — reformulating queries, summarising, simple follow-ups — and what needs a frontier cloud model
- Local work costs no cloud credits and never leaves the device
- Available on Pro, Max, and Enterprise plans; requires a recent Apple silicon Mac with enough memory for a 27B model
Why it matters
- Privacy: a meaningful share of what you ask never reaches a server
- Cost: Perplexity offloads compute to hardware you paid for, which is how it can raise limits without raising prices
- Speed: local steps skip the network round trip
- Precedent: expect ChatGPT, Claude, and Gemini desktop apps to follow; Apple Intelligence already works this way at the OS level
The catches
- A 27B model needs a lot of memory; older or 8 GB Macs will not qualify
- Which tasks go local is Perplexity's decision, not yours, and the boundary is not fully documented
- Battery and fan noise on laptops during heavy local inference
- Mac only at launch; Windows and mobile timelines are unannounced
The quiet significance is economic. Every AI company is paying for compute it would rather not; your laptop is compute they do not pay for. Hybrid is going to spread because it is good for privacy and very good for margins.
Related reading
Bottom line
Hybrid Compute is a small feature with a large future. For Perplexity Pro users on a capable Mac it is a free privacy and speed upgrade. For the industry it is the template every desktop AI app will copy within a year.
