Edge0 released on November 12 the Edge0-145B, an open-source artificial intelligence model with 145 billion parameters designed to run locally on a personal computer. Alongside it, the company also published the Edge0-120B, as well as its unified inference framework and deployment tools. According to Edge0, it is the first open 145-billion-parameter model built for on-device use.
The model is a sparse mixture of experts (MoE) derived from Qwen3.8-Flash-Next, with 512 experts and only about 10 billion parameters active per inference step. It works with context of up to 256,000 tokens and was trained natively in 4-bit, designed for local execution from the start rather than as a compressed version of a cloud model.
A 145B model on a laptop
According to figures published by Edge0 itself, the Edge0-145B runs on a compatible laptop with 29 GB peak memory, prefill above 1,100 tokens per second and generation above 80 tokens per second, with typical power consumption of around 120 W. The architecture shares large matrices across groups of 64 experts and uses a three-tier system — GPU memory, system memory and SSD — to load experts as needed.
For comparison, the family's smaller model, the Edge0-35B, runs on an iPhone 18 Pro Max at over 37 tokens per second using about 2.5 GB of memory.
Benchmarks: Edge0's own numbers
In tests published by the company, Edge0-145B scored 64.6 on DeepSWE 1.1, 86.52 on Terminal-Bench 2.1, 90.4 on GPQA-Diamond and 80.6 on IFBench. Edge0 says these results put it close to frontier cloud models in coding, reasoning and agent work. It is important to note that these figures are Edge0's own internal measurements and have not yet been independently verified.
In practice, the model was shown planning and completing full tasks on the computer: writing and debugging code, creating 3D scenes in Blender, producing an explainer video and generating playable games — all locally, with no calls to cloud models.
License and availability
The model weights are available on Hugging Face and the inference framework, called edge0-infer, is on GitHub, along with the Edge0 App. The model uses an Edge0 community license on top of the Qwen license: it allows research, internal use and commercial use of outputs, but selling the model or running it as a commercial service requires separate permission.
Running a model of this size on your own computer removes cloud per-token fees and keeps data and code on the machine, working even offline. The price is hardware: Edge0 itself notes that a compatible laptop with enough memory is needed and, for the larger versions, an NVIDIA GPU with CUDA 13.
Featured image: Edge0 chart showing the growth in size of open on-device models. Credit: Edge0 (official blog).
Sources: Edge0 — Edge0-145B: Open-Source On-Device AI with 145B Parameters; Daily AI Brief — Edge0 releases 145B open model built to run on one consumer GPU; GitHub — edge0-ai/edge0.
This text was produced with the assistance of artificial intelligence, cross-checking public posts and internet sources. Performance figures are measurements published by Edge0 itself and still await independent verification. Consult the original sources to confirm the information.
By Muse
