# Open Weight Models — Editorial Position v0.1

## Mission

Open Weight Models is an independent reference for the **verification and deployment reality** of open-weight AI.

Our goal is not to publish the biggest model list. Our goal is to make claims about open-weight models more useful, traceable and operational.

## Our thesis

Open weights are an infrastructure capability, not a marketing label.

Their strategic value is **optionality**:

- choice of infrastructure
- choice of runtime
- choice of provider
- choice of deployment geography
- ability to adapt the model
- a credible exit path from a single API vendor

We expect the future of AI to be hybrid. Managed proprietary APIs will remain valuable. Open-weight models will coexist with them and provide an alternative path when control, locality, customization, cost structure or portability matter.

## Our working definition

An open-weight model is an AI model whose trained parameters are made available so that others can obtain and run the model outside the original publisher's hosted API, subject to the model's license and technical requirements.

Open weight **does not automatically mean**:

- Open Source AI
- unrestricted commercial use
- public training data
- reproducible training
- secure deployment
- private deployment
- lower cost
- easy local inference

## Our sovereignty position

AI sovereignty is not autarky and it is not simply owning hardware.

We define practical sovereignty as **credible optionality and exit capability** across five layers:

1. Weight control
2. License freedom
3. Deployment control
4. Runtime portability
5. Exit capability

Open weights can strengthen sovereignty, but they do not remove dependencies on compute, chips, electricity, cloud capacity, software, data or engineering expertise.

## Editorial rules

- Exact checkpoint before family-level generalization.
- Exact license before “commercially usable.”
- Evidence label before checkmark.
- Primary source before secondary summary.
- Dated verification before “current.”
- Third-party benchmark remains third-party.
- OWM runtime-tested is reserved for runs physically reproduced by OWM.
- Unknown is an acceptable answer.
- No invented benchmark scores.
- No “open source” claim based only on downloadable weights.

## Why Open Weight Models exists

Model cards, licenses, runtimes, quantizations and deployment information are fragmented across model publishers, hubs, runtime projects and infrastructure providers.

OWM aims to turn that fragmentation into a source-first deployment reference with Model Passports, runtime evidence and change history.