Which API provider should you use with OpenCode?

Choose an OpenCode API provider by the models and billing arrangement you need. Zen offers a curated, usage-billed route; Go offers a subscription with defined allowances; direct provider accounts and gateways let you bring another API credential. There is no single cheapest choice for every coding workload.

This guide compares purchase and setup decisions, checked September 9, 2026. It is published by Ofox. It does not present an independent speed benchmark or claim that an Ofox account will fix upstream limits. If you already chose a provider, go straight to the OpenCode configuration guide.

Compare the purchase options

The Zen documentation describes its optional provider, credit billing and connection steps. The Go documentation gives the subscription price and limits. Treat the models included in Go as a specific catalog, not a subscription to every model OpenCode can connect to. Consult OpenRouter pricing and Ofox pricing documentation for their current terms.

Match the plan to your coding work

For occasional tasks, compare actual usage costs before committing to a recurring subscription. For daily work, check whether a subscription’s included models and rolling limits cover your busy periods. A monthly allowance alone does not tell you whether a long session can continue.

For a project that depends on a specific model, choose model and protocol compatibility first. A model that answers a short prompt may still behave differently on tool calls, structured results or a long conversation. For a team, also inspect key permissions, spend controls and the provider’s data terms before using private repositories.

A useful cost measure is total billed cost per accepted coding task. Include repeated attempts and the tool conversations needed to reach a reviewed result. The lowest output-token rate can be a poor bargain if the task requires substantially more attempts; this is a measurement method, not a claim about any provider in the table.

Evaluate Ofox with one project first

Start with the Ofox model catalog and select a model whose documented protocol and tool capabilities fit OpenCode. Ofox publishes an OpenAI-compatible base URL of https://api.ofox.io/v1 in its SDK documentation. That URL does not mean every model uses the same runtime or request schema.

To try that connection, create an Ofox account, then open API Keys. Check billing and model access before sending a task. This is an account-creation path, not an offer of free inference.

For the regular OpenCode configuration documented at /docs/providers/:

1. Run opencode --version and use the documentation for that installation.
2. In OpenCode, run /connect, choose Other, and register your Ofox key under a provider ID such as ofox.
3. Declare the same provider ID and the chosen model in opencode.json. Use the correct runtime for the actual protocol; do not paste an API key into the file.
4. Select the model with /models, then evaluate a small coding task in a disposable project.

The configuration guide contains the JSON template and explains the regular and v2 field differences. Use that version-specific guide rather than combining older TOML examples with the current JSON schema. The official Providers documentation is the authority for OpenCode’s configuration format.

Decide after a small, comparable evaluation

Use the same repository snapshot, task instructions and acceptance tests for each option. Record model ID, provider, total charge, task completion time, tool-call failures and whether the reviewed result passed. Keep model quality and provider delivery separate: changing both at once cannot isolate the cause of a difference.

If your current account is on OpenRouter, use the Ofox migration checklist to inspect routing fields and model mappings before switching. No paid cross-provider coding benchmark was run for this comparison.