Last updated: 11/18/2026Last verified: 2026-11-13
Edge0 is an open-source on-device AI stack: open-weight Edge0 language models (145B and 120B for a single NVIDIA GPU, 35B and 8B for phones) plus the edge0-infer engine with a CLI, an OpenAI-compatible local server and the Edge0 Chat desktop app, so chat, coding and agent work can run locally without per-token cloud fees.
What is Edge0?
Edge0 is an open-source on-device AI stack for running chat, coding and agent work on your own hardware. It pairs open-weight Edge0 language models with the edge0-infer engine, which provides a CLI, an OpenAI-compatible local server and the Edge0 Chat desktop app. The latest Edge0-145B and Edge0-120B models (released November 2026) are latent mixture-of-experts models derived from Qwen3.8-Flash-Next and trained natively at 4 bits; they run on a single NVIDIA GPU under Linux or Windows. Local inference means no per-token cloud API fees, though hardware and electricity still cost money.
Key Features
- Open-weight Edge0-145B and Edge0-120B models on Hugging Face in Q3_K and NVFP4 builds.
- Smaller Edge0-35B and Edge0-8B checkpoints aimed at phones; the README says native-engine support for 8B is still planned.
- edge0-infer CLI with run, chat, bench, serve and info commands.
- An OpenAI-compatible local server that also answers POST /v1/messages (Anthropic Messages API) for native models.
- Edge0 Chat desktop app: release 0.2.0 ships Windows x64 and Windows Arm64 packages; model weights are downloaded separately.
- Expert weights are streamed across GPU memory, system memory and SSD so large sparse models fit consumer hardware.
- Edge0-35B runs on macOS (Metal), iOS (Metal), Android (CPU), Linux (CPU) and Windows (CPU or Vulkan), per the GitHub README.
Best For
Pricing
Edge0 is free to download and run. The inference code is Apache-2.0. The model weights use the Edge0 Community Model License 1.0 (12 November 2026), which allows personal, research and internal business use and commercial use of outputs at no charge. Selling access to the model, running a paid inference or fine-tuning service, or selling an independent AI coding or office assistant built on it requires a separate written commercial license, and the upstream Qwen license still applies. Checked 2026-11-13.
Pros & Cons
Pros
- Open weights plus open-source inference code (Apache-2.0).
- Covers both a CLI and a desktop chat app.
- OpenAI-compatible and Anthropic Messages-compatible local server.
- No per-token cloud fees for local use.
- Model range from phone-sized checkpoints up to 145B.
Cons
- Edge0-145B and Edge0-120B need Linux or Windows with an NVIDIA GPU and CUDA 13; the 0.2.0 release notes ask for compute capability 7.5 or newer, at least 16 GiB of GPU memory and 20 GiB of system memory.
- The README platform table (read 2026-11-13) lists no macOS, iOS or Android support for the 145B and 120B models, although the launch blog mentions a MacBook Pro and Apple M-series chips.
- The 0.2.0 Edge0 Chat packages are Windows-only.
- Commercial model offerings need a separate license from Edge0.
- Published benchmark and speed figures come from Edge0’s own internal evaluations; this overview did not run the models.
Alternatives
Google’s app for running smaller open models on phones, tablets or a Mac.
Checks which local LLMs fit your RAM, CPU and GPU before you download one.
Hosted chat and API access to DeepSeek models instead of running weights locally.
FAQ
Details
Platform
Features
- Open-weight Edge0-145B and Edge0-120B latent mixture-of-experts models on Hugging Face in Q3_K and NVFP4 builds, derived from Qwen3.8-Flash-Next and trained natively at 4 bits
- edge0-infer engine with run, chat, bench, serve and info commands, an OpenAI-compatible local server, and POST /v1/messages (Anthropic Messages API) support for native models
- Edge0 Chat desktop app (release 0.2.0, Windows x64 and Arm64 packages) for chatting with local models; the blog says core capabilities work offline without an account or usage quota
- Streams expert weights across GPU memory, system memory and SSD so large sparse models fit consumer hardware; earlier Edge0-35B runs on macOS (Metal), iOS (Metal), Android (CPU), Linux and Windows
Languages
Known limitations
- Edge0-145B and Edge0-120B currently need Linux or Windows with an NVIDIA GPU and CUDA 13; the 0.2.0 release notes ask for compute capability 7.5 or newer, at least 16 GiB of GPU memory and 20 GiB of system memory. The GitHub README platform table (read 2026-11-13) lists no macOS, iOS or Android support for these two models, even though the launch blog mentions a MacBook Pro and Apple M-series chips; check the README before planning a Mac setup
- The 0.2.0 Edge0 Chat packages are Windows-only and model weights are not included; you download the 145B/120B checkpoints from Hugging Face separately
- The model weights use the Edge0 Community Model License 1.0 (12 November 2026): personal, research, internal business use and commercial use of outputs are allowed at no charge, but selling model access, running a paid inference/fine-tuning service, or selling an independent AI coding or office assistant built on the model needs a separate written commercial license, and Qwen upstream license terms still apply. The inference code is Apache-2.0
- Benchmark and speed figures (for example 29GB peak memory and 80+ tokens/s on a laptop) come from Edge0’s own internal evaluations; this listing did not run the models







