ModelBrief · Edge0-145B · Topic Page
Last updated Nov 20, 2026

Edge0-145B License and Cost: Weights Are Not Open Source

ModelBrief · Published 2026-11-20 · Fact-check 2026-11-20 · Reference: Hugging Face model cards, Edge0 official blog, GitHub Edge0-AI/Edge0

After the Edge0-145B launch, the licence — not the benchmarks — is what gets misreported most often. The official blog used the word "open-sourcing", and several launch write-ups followed it with "open source". But the weights are actually governed by a custom community licence. The cost of getting this wrong is real: building a commercial service on the assumption of "open source" could mean infringing without a written licence. This article works through the licence texts clause by clause — the licence text is the November 12, 2026 version; check the original on the Hugging Face model card before relying on it.

Three licences, each covering its own thing

SubjectLicenceNature
Model weights (Edge0-145B)Edge0 Community Model License 1.0 (Nov 12, 2026)Custom community licence, not an open-source licence
Base-model parts (Qwen3.8-Flash-Next derivative)Qwen Community License 1.0Applies as well; the release-scope file states this is not evidence of a separate Qwen commercial authorization
Framework and deployment code (edge0-infer / edge0-strata)Apache 2.0A genuine open-source licence — it covers the code, not the weights

The most common confusion starts here: the GitHub repository header says Apache 2.0, and some write-ups conclude "Edge0-145B is Apache 2.0 open source". Wrong. Apache 2.0 covers only the edge0-infer and edge0-strata code; the weights have their own licence. The code is open source; the weights are open but restricted — write them separately.

The weights licence: what is free, and what needs a separate written licence

Free of charge: personal use, research, modification, non-commercial sharing, internal business use, and commercial use of generated outputs — all free. Running the model locally for personal use, doing research, modifying the weights, using it inside a business, and selling what the model generates are all inside the free scope.

Needs a separate written licence: selling, renting or licensing the weights themselves; operating a commercial model service for third parties (fees, subscriptions, ads or sponsorship, where customers control inference inputs); offering an independent commercial coding or office-productivity product, hosted or on-device. In short: using it yourself is free; offering it to others as a service or selling a product built on it requires a written licence from Edge0 first. The release-scope file adds that this licence is not evidence of a Qwen commercial authorization — the base model's commercial rights are a separate question.

One-line summary: Edge0-145B is "open weights", not "open source". Personal and research use is nearly free; commercial services and commercial products have a clear threshold. The vendor blog's "open-sourcing" wording does not match this, and this site does not repeat it.

Price: no per-token billing, only local cost

Edge0-145B has no hosted API and no per-token price: OpenRouter does not list the model, and there is no cloud endpoint. Edge0's own framing is "no corresponding cloud-model token or API fees". The cost has two items only: hardware and electricity.

Cost itemEdge0-145B terms
API callsNone (no hosted API, no OpenRouter listing)
Weight downloadFree (open weights on Hugging Face)
Framework and toolsFree (Apache 2.0)
HardwareOne NVIDIA GPU with 16GB+ VRAM (compute capability 7.5+), 20 GiB+ system RAM, CUDA 13; measured whole-machine demand about 61.5 GiB (NVFP4) / 54.4 GiB (Q3_K) of VRAM + host RAM, plus about 40–48 GiB of disk
ElectricityAbout 120W typical power (vendor figure); Edge0 App core capabilities work offline, with no account and no usage quota

"Zero token cost" is not "zero cost"

The official pitch of "zero token cost" is true — but it tells half the story: the marginal cost is zero, the fixed cost is not. Before running Edge0-145B you need a machine that meets the bar — a 16GB+ VRAM GPU, 20 GiB+ system RAM, CUDA 13, and about 40–48 GiB of disk; measured whole-machine memory demand runs about double the headline 29GB (about 61.5 GiB for NVFP4). For someone who already owns that machine, the marginal cost is indeed only electricity (about 120W typical power); for someone who does not, the first bill is hardware, not tokens. Compare it with cloud API models this way: cloud bills per use with zero fixed cost; Edge0-145B pays its fixed cost once and only electricity afterwards — the heavier your usage, the better the local route looks; light usage is cheaper in the cloud.

One more note: the Edge0 App's "free" means its core capabilities work offline with no account and no usage quota — not that every feature is free. Commercial services and commercial products remain restricted by the weights licence, as above.

ModelBrief's take: The licence is the part of the Edge0-145B story most worth its own article: personal and research use is almost entirely free, a real advantage over closed models; but the word "open-source" is wrong — the weights are open weights under a custom licence, commercial services and products need a separate written licence, and the code's Apache 2.0 does not extend to the weights. The cost ledger is the same: no token bill is real, but the hardware bar and the power bill are a separate ledger — check the about-61.5 GiB whole-machine memory demand before installing.