Skip to content

AI Models Used by OpsDev.nz

OpsDev.nz uses AI assistance for software development through the OpenCode editor. Model access is provided via OpenCode Zen and OpenCode Go subscriptions.

The current set of models available through the OpenCode Go subscription is maintained by the provider:

How it works

OpenCode is not the model owner. It is a broker. When we prompt a model (Grok, GPT, Qwen, Kimi, DeepSeek, GLM, and others), OpenCode forwards the request to that model's provider and streams the response back. The actual compute — the model inference — runs on each provider's own infrastructure, not on OpenCode's servers and not on our machines.

OpenCode pays the providers per token for that inference; it does not pay for (or perform) model training. Our Go/Zen subscriptions simply resell access to those provider APIs under a flat monthly fee.

How we use the models

  • Task-driven selection. We don't pin one model. We bounce between them depending on the job (code generation, documentation, spec maintenance, infrastructure automation across the OpsDev.nz open-source toolkit).
  • Inference only. We consume model outputs; we do not train or fine-tune models, so our footprint is limited to inference compute on the providers' infrastructure.
  • Many are open-weight. A large share of the available models (Qwen, DeepSeek, GLM, Kimi, and others) are open-weight, which tends to be offered as a cheaper, more efficient commodity than proprietary alternatives.

Sustainability notes

Using AI assistance consumes energy in the model providers' data centres, and that use is ours. We choose to run it. We will endeavour to learn more about the sustainability practices of our hosting and model providers. OpenCode gives us the flexibility to switch between models, so where possible we will favour providers whose practices align with our low-impact, open-source values.

Our full position is in the sustainability statement, and our machine-readable disclosure is in carbon.txt.