86
of 2025's 107 notable AI models had unreleased training code; 4 were fully open source
What the chart shows
Even when model weights are shared, the code used to train a model usually is not. In 2025, 86 of 107 notable models had unreleased training code and 4 had open-source training code; for 17 the status is unknown.
Open-source training code was more common in 2023, when 30 notable models released it.
Key numbers
| Measure | Value |
|---|---|
| Unreleased training code, 2025 | 86 |
| Open-source training code, 2025 | 4 |
| Unknown, 2025 | 17 |
| Open-source training code, 2023 | 30 |
How to read it
- Open with restrictions combines non-commercial and restricted-use licences.
- Training code is different from model weights: see the related chart for how the models themselves are released.
Source: Epoch AI, Data on AI models (CC BY 4.0), accessed 19 October 2026. Analysis and chart by AI Adoption, following the method of the AI Index Report 2026 (Stanford HAI), figure 1.1.9.
Numbers checked against the published data on 19 October 2026.
Download the data (CSV) · Cite this chart
Numbers checked against the published data on 19 October 2026.
Download the data (CSV) · Cite this chart
Discussion