Open, accessible,
correctable AI.
For the next decade of small language models.
OpenSLM.ai is the home for the Open Small Models Accord and related work on open, accessible, and correctable AI. We document principles, build tooling, and publish research aimed at making small language models genuinely open: inspectable, reproducible, and correctable by the people they affect.
- LLogic
- WWeights
- DData
- RRepresentation
The principles
Ten principles for open, small AI.
The accord names ten principles. Read each as a single idea, then read the full text for the argument behind it.
-
Small models are where openness, accessibility, and accountability can actually meet.
-
Openness is a layered property; disclosure follows the layer.
-
Open models are built at the scope of a specific domain, task, and language.
-
Running a model is not the same as reproducing it.
-
Local inference keeps accountability with the people affected.
-
Safety comes from inspectable architecture, not from closure.
-
Agents are systems, and the system is what requires openness.
-
Evaluation must measure what deployment requires, not what is convenient to report.
-
Coordination through open standards bodies matters now, while the choices remain contestable.
-
Accountability requires visible disclosure, community contestation, and revisable commitments.
Initiative
Built in the open.
OpenSLM.ai grows with the work. The accord is the founding artifact; the initiative reaches further — into tooling, evaluation, and infrastructure for small models that can be inspected, reproduced, and corrected.