Glossary
open weights
A model release that publishes the trained parameters under some downloadable license, distinct from "open source" which (per OSAID) also requires data and training-code openness.
The 2024 to 2026 term-of-art for a model released as a downloadable weight file but without the training data, training code, or both. Most LlamaweightsMeta's open-weight model family, the most widely deployed open release through 2024 to 2026, released under the source-available Community License with an MAU cap and acceptable-use clause. Open full entry , MistralweightsA French open-weight model family from Mistral AI, released mostly under Apache 2.0 with strong performance per parameter and notable MoE variants (Mixtral, Mixtral 8x22B). Open full entry , QwenweightsAlibaba's open-weight model family, leading the multilingual and Chinese-language open-weight space, released under Apache 2.0 with sizes from 0.6B to 235B parameters. Open full entry , DeepSeekweightsA Chinese open-weight family known for the V3 MoE base model and the R1 reasoning model, both released under permissive licenses and unusually transparent in their training-cost reporting. Open full entry , and GemmaweightsGoogle's open-weight model family derived from Gemini research, with source-available licensing that includes an acceptable-use clause and license-revocation hook. Open full entry releases are open- weight: the parameters are public, the rest is not. Under the OSAIDgovernanceThe OSI's October 2024 definition of "open source AI," requiring not just weights but enough information about data, code, and architecture for third parties to reproduce the system. Open full entry definition this does not qualify as open-source AI.
The distinction matters in practice. With open weights you can run, fine-tuningtrainingContinued training of a pretrained base model on a smaller, task-specific dataset to specialize its behavior without retraining from scratch. Open full entry , and inspect the model. You cannot reproduce the training run, audit what the model was trained on, or replicate the safety and quality choices made during post-trainingtrainingEverything that happens after pretraining ends: supervised fine-tuning, preference optimization, red-teaming, distillation, and safety work that turns a base into a shippable assistant. Open full entry . For most research and deployment purposes this is acceptable; for legal compliance, audit, or scientific reproducibility it often is not.
The label has come to function as a useful compromise term. Saying “open weights” is honest about what was and was not released, where “open source” tends to overclaim. The OSIgovernanceThe nonprofit that maintains the canonical Open Source Definition for software since 1998, and the OSAID definition for AI as of 2024. Open full entry ’s position is that “open weights” is fine as a description; it is the upgrade to “open source AI” that requires more.