What is an open-weights model? Open weights vs open source
Also called open-weight model, open model, open-weight LLM
Definition
An open-weights model is an AI model whose trained parameters are published for download, so anyone can run, fine-tune, or quantize it on their own hardware under the terms of its license.
An open-weights model is a model whose trained parameters, the weights, are published for anyone to download. A team can run it on its own GPUs, quantize it to fit smaller hardware, or fine-tune it on private data that never leaves its infrastructure. Whoever downloads the weights also takes on serving, scaling, and upgrades. The release tracker marks each model release as open-weights or proprietary and can filter to open weights only.
Open weights vs open source#
Open weights means the parameters can be downloaded. Open source AI, as the Open Source Initiative (OSI) defines it, also requires the training code and enough information about the training data to rebuild the system. Most open-weights releases publish only weights and inference code, often under licenses that restrict some uses.
The Open Source AI Definition is at version 1.0, released by the OSI on October 28, 2024. It requires three components under open terms:
- Data information: a complete description of the training data, detailed enough that a skilled person could build a substantially equivalent system, plus where to obtain the public and third-party data.
- Code: the complete source code used to train and run the system, including data processing and filtering.
- Parameters: the weights and other configuration settings, which can include checkpoints from key stages of training.
The definition asks for a description of the data, not the data itself. In the OSI's validation phase, Pythia, OLMo, Amber, CrystalCoder, and T5 passed, while Llama 2, Grok, Phi-2, and Mixtral did not, because components were missing or their legal terms conflicted with open source principles. The OSI states that these results are not certifications.
Open-weights licenses#
Open-weights licenses fall into two broad groups. Standard permissive licenses such as Apache 2.0 or MIT allow commercial use, modification, and redistribution with few conditions. Custom licenses written by the model developer can add acceptable use policies, non-commercial terms, or limits for very large companies, and they differ from one developer to the next.
Two releases show the range. OpenAI published gpt-oss-120b and gpt-oss-20b under Apache 2.0 in August 2025. Meta released Llama 3.1 under the Llama 3.1 Community License, which incorporates an Acceptable Use Policy and requires any licensee whose products had more than 700 million monthly active users to request a separate license from Meta. Read the license for the exact model version you deploy, since terms can change between releases.
Running an open-weights model#
Running an open-weights model means choosing hardware, a numeric precision, and a serving engine. Engines such as vLLM, SGLang, llama.cpp, and Ollama load the weights and serve requests; quantized files in formats such as GGUF cut memory needs enough to run smaller models on a laptop.
Teams that want the model without managing GPUs can use inference providers that host popular open-weights models behind an API.
vLLM vs Ollama compares the serving engines most teams use to run open-weights models.
Related terms
Guides that cover open-weights model
- What is AI engineering? The discipline, the stack, and the work
- Reasoning models: how they work and when to use them
- LLM benchmarks explained: what they measure and how to read them
- vLLM vs Ollama vs llama.cpp vs SGLang: which engine to use
- Best LLM for coding: models compared on agentic coding benchmarks
Cite this definition
The AI Engineering Editors. "What is an open-weights model? Open weights vs open source." The AI Engineering, updated Nov 06, 2026. https://theaiengineering.com/glossary/open-weights-model
- Title
- What is an open-weights model? Open weights vs open source
- Author
- The AI Engineering Editors
- Publisher
- The AI Engineering
- Updated