What Is Open-Weight Model?
An open-weight model is an AI model whose trained weights are published for anyone to download and run, so it can be self-hosted, fine-tuned or served by a choice of providers, unlike closed models that are only available through the developer's own API.
Also known as:
Open-Weight vs Open Source
Open weights means the trained parameters are available. It does not necessarily mean the training data, training code or a permissive license are available too. Licenses vary, and some limit commercial use, so it is worth checking each model's terms before relying on it.
Why Teams Use Open-Weight Models
- Cost. Open-weight models are often served at a fraction of the price of frontier models.
- Control. They can run in your own environment, which can help with data residency and privacy requirements.
- Choice of provider. The same model can be served by several providers, which reduces lock-in.
- Customization. Weights can be fine-tuned for a narrow task.
Limitations
The strongest open-weight models often trail the leading closed models on the hardest reasoning tasks. Running them yourself adds infrastructure and operations work, and the security of a self-hosted model is your responsibility, including the provenance of the weights you download.
Where Open-Weight Models Fit in an AI Harness
Most AI pipelines include many tasks that do not need the most capable model, such as screening, classification and evidence gathering. As open-weight models have become able to run reliable multi-turn tool loops, they have become a practical choice for those stages, with frontier models reserved for the hardest reasoning.
How Nullify Uses a Mix of Models
Nullify's harness runs a mix of models and routes work to cheaper ones when that is all a task needs. By mid-2026, open-weight models could run reliable multi-turn tool loops at roughly a tenth of frontier token prices, and Nullify rebuilt a triage screening stage around a cheaper model that investigates the repository itself. See how Nullify triages findings.
Key Takeaways
- Open-weight models publish their trained weights so anyone can run them.
- Open weights is not the same as open source, so check each license.
- They offer lower cost, more control and less lock-in, with some capability trade-offs.
- They fit well in screening and evidence-gathering stages of an AI harness.
Read More About Open-Weight Model
Triage & Remediation
See Nullify in Action
Nullify routes each task to the right model, so you are not paying frontier prices for every step.