# How open is this AI?

HTML tool: https://corelooppress.com/tools/open-or-closed/

Eight short scenarios distinguish access, use rights, disclosure and practical deployment. A quiz result counts which distinctions the reader recognized; it is not a model-openness score or an assessment of a company. Choices stay in the browser. The page links its primary definitions and examples, checked September 27, 2026.

The share image is a four-question field note for choosing a model. It leads with the distinction between free access and openness; the reader's quiz score is a small personal detail.

- **Access:** A free API is not a copy of the weights. Downloaded weights also need compatible files, compute and suitable terms for self-hosting.
- **Rights:** Downloadable weights do not by themselves satisfy the Open Source AI Definition. Read the terms governing the exact release, including any incorporated policy.
- **Disclosure:** A license does not publish a training recipe. OSI requires sufficiently detailed data information, not a public copy of every raw training example.
- **Practical use:** Hosting an open-weight model has compute and operational costs. Keeping prompts in one's own environment changes the data path, but privacy still depends on controls and practices.

The quiz uses OpenAI's separately published hosted and downloadable models as one illustration and Meta's Llama 4 license as an example of release-specific conditions. It does not rank models or interpret a license for a particular user's circumstances.

Primary sources: [OSI Definition 1.0](https://opensource.org/ai/open-source-ai-definition), [OSI FAQ](https://opensource.org/ai/faq), [OpenAI open models](https://openai.com/open-models/), [OpenAI GPT-6 Astra availability](https://openai.com/index/gpt-6-astra/), [Meta Llama 4 license](https://github.com/meta-llama/llama-models/blob/main/models/llama4/LICENSE), [Hugging Face endpoint pricing](https://huggingface.co/docs/inference-endpoints/pricing), and [NIST Privacy Framework](https://www.nist.gov/privacy-framework).