Frequently
asked questions

1. Q.01

   What is the Open Small Models Accord?

   The Open Small Models Accord is a standards-style document setting out principles for open, accessible, and correctable AI, with a focus on small language models. It is hosted at openslm.ai and released under CC0 1.0.

   The accord defines four layers of openness (Logic, Weights, Data, and Representation, together called LWD-R) and asks signatories to make visible which layers their releases meet, and to move toward fuller openness over time. It is a direction of travel, not a fixed bar.

2. Q.02

   Who can endorse it?

   Researchers, model developers, infrastructure providers, civil society organizations, enterprises, and standards bodies. Individuals can also endorse.

   Endorsement is a statement of direction, not a claim that a specific release meets every layer of LWD-R today. Signatories commit to publishing a public disclosure of which layers their releases meet, and to working toward closing the gaps over time.

3. Q.03

   Is endorsement legally binding?

   No. Endorsement is a public commitment to the principles, not a contract. The accord is released under CC0 1.0, which imposes no obligations and grants no rights beyond what is already in the public domain.

   Accountability comes from the public LWD-R disclosure each signatory publishes, which any third party can compare against the signatory’s actual releases. The community can contest endorsements that become symbolic without corresponding practice.

4. Q.04

   How does this relate to the OSI Open Source AI Definition?

   The accord builds on the OSI Open Source AI Definition rather than replacing it. The OSI definition specifies what counts as open AI in the legal and licensing sense, particularly for code, parameters, and training data information. The accord draws on that work and extends it by adding Representation as a fourth layer.

   The accord also treats openness as a direction of travel rather than a binary test. A release can be assessed layer by layer (L, W, D, R), with the gaps disclosed and the path to closing them documented. The OSI definition remains the reference for what each layer’s openness means in licensing terms.

5. Q.05

   Can I translate, adapt, or republish the accord?

   Yes. The accord is released under CC0 1.0, which places it in the public domain worldwide. Translation, adaptation, and republication are encouraged.

   We ask, but do not require, that translations:

   • link back to the canonical version at openslm.ai
   • indicate which accord version was translated (the current canonical version is shown in the Attribution block)
   • credit the translator

   To contribute a translation back to this site so others can find it, see the Translations page.

6. Q.06

   What if my release does not meet all four LWD-R layers?

   That is expected. Few releases in 2026 meet all four layers fully. The accord asks signatories to publish a public LWD-R disclosure that says which layers their release meets, where the gaps are, and what they intend to do about them.

   Endorsement is a direction of travel, not a claim of arrival. A signatory whose current release only meets two layers but is moving toward all four is exactly the case the accord is designed for.

7. Q.07

   How do I contribute changes to the accord?

   Open a pull request at openslm-ai/website.

   Editorial changes (typos, clarity, formatting) are welcome at any time. Substantive changes to the principles go through a public review process before being released as a new version. The accord is versioned, and revisions are released openly.

8. Q.08

   Who maintains the accord?

   The accord was authored and initiated by Anivar Aravind and is hosted at openslm.ai.

   As endorsement opens, a maintainer body composed of signatories will take on revision, governance, and translation oversight. Until then, editorial decisions are made in the open via the public repository at openslm-ai/website.

Question we have not answered? Open an issue at openslm-ai/website or write to hello@openslm.ai.