What happened
Netlify announced “full model freedom” across two of its production AI tools on August 6. Through partnerships with OpenRouter (a marketplace that routes requests to many different AI models through one connection) and OpenCode (a free, publicly available coding agent that can be pointed at different underlying models), Netlify now offers open-weight models alongside Claude, GPT, and Gemini. Open-weight models are AI models whose underlying files are published so anyone can run or host them, instead of being locked behind one company’s paid service. DeepSeek, Qwen, GLM, and Kimi, all named in Netlify’s announcement, fall into that category.
This lands in two places. AI Gateway is Netlify’s routing layer for AI calls inside apps you’ve built: it sits between your app and whichever model provider you pick, so you can change the underlying model without rewriting the code that connects your app to that provider. Agent Runners are Netlify’s own in-dashboard coding agents: you open a project, describe a change, and an agent edits the code and builds a preview version of your site that you can check before it replaces the live one. Until now, Agent Runners let you choose from four engines: Claude Code, OpenAI Codex, Google Gemini, or OpenCode. The open-weight models arrive specifically through OpenCode, which you select as the agent and then point at a model like DeepSeek or Kimi K3 instead of a closed one.
Why it matters
Model choice affects both what a task costs and how well it gets done, and different models are better at different kinds of work. Being able to swap the model behind an agent, or behind a live request your app sends out, without touching the connecting code, is a genuine convenience. It also means a bad day for one model provider, an outage or a sudden price change, doesn’t have to become your app’s bad day.
The gap is pricing. Netlify’s announcement doesn’t publish what an open-weight model run costs compared to a Claude or GPT run. Netlify’s own Agent Runners documentation confirms that runs are billed in credits, on Free, Personal, and Pro plans, covering AI inference and compute time. It does not list per-model credit costs. So “open-weight models are cheaper” is the industry’s general reputation for these models, not a number Netlify has committed to for this feature. Don’t assume a DeepSeek or Kimi run will cost less in credits than a Claude run until you’ve checked.
There’s also a data-handling question worth a beat of caution. Open-weight models on Netlify are hosted through the OpenRouter partnership, a third party, not run directly by Netlify or by the model’s original lab. If your app handles user data you’re accountable for, check OpenRouter’s and the specific model’s data retention policy, how long they keep the data you send and what they do with it, before pointing your app’s AI Gateway traffic at a new model. The terms may differ from whatever provider you were using before.
Who should care
Anyone using Netlify’s Agent Runners for day-to-day code changes, and anyone using AI Gateway to route AI calls inside a deployed app who wants a cheaper or alternate model without rewriting how their app talks to its model provider.
What builders should do next
In your Netlify project dashboard, open the “Build with an AI agent” section and start a new Agent Run. Choose OpenCode as the agent, then pick an open-weight model like DeepSeek or Kimi K3 from the model list, and run a small bounded task, such as fixing a failing test or adding input validation to a form. Run the identical task again with Claude Code or Codex as the agent. Compare two things: the credits each run consumed, visible in the agent run detail view, and whether the output passed the same tests without you needing to fix follow-up issues.
That comparison, not the announcement itself, is what tells you whether switching is worth it for your project.
End of article