An open-weight model has its parameters published for download, so it can run on your own infrastructure or through a hosting provider. Open-weight models such as Llama and DeepSeek-class models are often cheaper per token, which makes them a common target for cost-optimized routing.
Why it matters
Open weights remove vendor lock-in and often lower the per-token price, because the model can be served by whoever runs it most efficiently.
How it works
The model's parameters are published, so a hosting provider - or your own infrastructure - can serve it. Through a gateway, an open-weight model is just another catalog entry you can route to or fall back on.