Running a large, capable model on your own laptop has meant either a small model or a very expensive machine. Edge0, whose 35B phone model we have reviewed, now claims a 145-billion-parameter model that runs on a suitably equipped laptop at a measured 29GB peak memory, with weights and inference code released openly. If the numbers hold, it moves serious local coding agents onto hardware many developers already own.
The design explains how the memory figure is possible. Edge0-145B is a mixture-of-experts model in which each group of 64 experts shares two large matrices, with each expert keeping only a small set of its own parameters; about 10B parameters are active per step. The engine spreads the model across three tiers: frequently used experts stay in GPU memory, an expert library of about 40GB sits in system memory, and about 48GB of vocabulary data lives on the SSD and is read on demand. The CPU and GPU work in parallel, and speculative decoding checks several candidate tokens at once. Edge0 reports more than 80 tokens per second.
The benchmark claims are strong. Edge0 reports 64.6 on DeepSWE 1.1, 86.52 on Terminal-Bench 2.1, 90.4 on GPQA-Diamond and 80.6 on IFBench, and a mean of 72.8 across six benchmarks against 71.7 for Qwen3.8-Flash-Next, whose standard 4-bit deployment it estimates at 70GB. That puts Edge0 at about 40% of that memory and makes a 48GB laptop the practical floor. Day-one support covers macOS, Windows and Linux, including Apple M-series chips and NVIDIA RTX cards.
For anyone running local models through llama.cpp, Ollama or LM Studio, the interesting part is the streaming design rather than the parameter count. Keeping most experts off the GPU and reading vocabulary from the SSD is a different trade from simply quantising harder, and it depends on fast storage. Edge0 also says the full stack is released so results can be reproduced.
The doubts are significant. Edge0's own results for Qwen, DeepSeek and Claude come from its internal evaluations, the competitor memory figures are estimates not yet measured, and the technical report with the full setup is still to come. The speed and memory numbers come from Edge0's own testing on unspecified laptop hardware. Our review of its 35B release found careful engineering with nearly every speed figure supplied by its makers, and that caution applies here until independent runs appear.