Home › AI models explained › Open source vs open weight LLMs

Open source vs open weight LLMs: what the licences allow

"Open" is used loosely for AI models. Some are open source by a formal definition; many more are open weight; some are neither. This page explains the difference using the Open Source Initiative's definition and two real examples, and gives a checklist for reading a model licence before you build on it. Section hub: AI models explained; for the legal angle on AI see AI law, copyright and safety.

Quick answerOpen weight means a model's parameters are published. Open source AI, per the Open Source Initiative's 1.0 definition, also needs sufficiently detailed training-data information and the complete training and running code, all under OSI-approved terms. Always read the licence: examples include Apache 2.0 (gpt-oss) and MIT (Whisper).

Prices last verified against official provider pages on November 13, 2026. How we verify prices. Written by the Chelure Enterprises Editorial Team.

Two meanings of "open"

	Open weight	Open source AI (OSI Definition 1.0)
What is released	The model's trained parameters (weights)	Parameters, plus the complete training and running code, plus sufficiently detailed information about the training data
Licence	Whatever the publisher chooses (for example Apache 2.0, MIT, or a custom licence)	Code, data information and parameters under OSI-approved terms
Freedoms	Defined by the licence	Use for any purpose, study, modify (including outputs) and share, with or without changes
Can you reproduce training?	Usually not	The definition aims for enough information for a skilled person to build a substantially equivalent system

The Open Source AI Definition

The Open Source Initiative's Open Source AI Definition 1.0 lists four freedoms and three components needed to exercise them. Data Information means sufficiently detailed information about the data used to train the system, so that a skilled person can build a substantially equivalent system, made available under OSI-approved terms. Code is the complete source code used to train and run the system, representing the full specification of how the data was processed and filtered and how training was done, under OSI-approved licences. Parameters are the model weights or other configuration settings, made available under OSI-approved terms.

Examples

Model	What the publisher says	Source
gpt-oss-120b and gpt-oss-20b (OpenAI)	Open-weight language models available under the Apache 2.0 licence, announced August 5, 2025	OpenAI announcement
Whisper (OpenAI)	Speech recognition model; code and model weights released under the MIT licence	Whisper repository README

Both examples use permissive licences for the weights. Neither announcement says the training data is documented to the level the Open Source AI Definition requires, and we have not assessed them against the definition. We could not open Meta's Llama licence page from our systems, so we do not summarise it; if you plan to use Llama models, read that licence yourself, since custom licences can include conditions beyond Apache or MIT.

Before you build on a model: a licence checklist

Which licence covers the weights, and which covers the code? They can differ.
Commercial use: permitted without conditions, or limited by size, revenue or user numbers?
Use restrictions: does an acceptable-use policy limit what you can do?
Attribution and naming: must you credit the model or keep a notice?
Outputs and derived models: can you use outputs to train or improve other models?
Hosting cost: running weights yourself replaces per-token fees with hardware cost; compare with API prices in the price comparison.

Frequently asked questions

What is the difference between open source and open weight?

Open weight means a model's trained parameters (weights) are published so you can run and often fine-tune it. Open source AI, as the Open Source Initiative defines it, asks for more: sufficiently detailed information about the training data, the complete source code used to train and run the system, and the parameters, all under OSI-approved terms.

Is gpt-oss open source?

OpenAI calls gpt-oss-120b and gpt-oss-20b open-weight models and says they are available under the Apache 2.0 licence. OpenAI's announcement does not claim they meet the Open Source AI Definition, and we have not assessed whether they do.

Is Whisper open source?

OpenAI's Whisper repository releases code and model weights under the MIT licence. Whether the training data is described in enough detail to meet the Open Source AI Definition is a separate question that we have not assessed.

What are the four freedoms?

The Open Source AI Definition 1.0 says an open source AI system gives you the freedom to use it for any purpose, study how it works, modify it (including its outputs) and share it, with or without changes.

Can I use an open-weight model commercially?

It depends entirely on its licence. Apache 2.0 and MIT are permissive; some open-weight models ship with custom licences carrying extra conditions, so read the licence of each model before building on it.

Sources

Primary sources opened on November 13, 2026. Prices come from our verified price table (see the methodology).