OpenCode pricing: Go, Zen and your own key, and what a call really costs

OpenCode is an open-source coding agent under the MIT licence, and it costs nothing. What costs money is the model it runs, and OpenCode gives you three ways to pay for it: a subscription, a pay-as-you-go gateway, or a key from a provider you already use. Every price here was read from OpenCode's own docs on 20 October 2026; they change often, so check the page before you budget.

Is OpenCode free?

Yes. The agent is free and open source, and it runs a few free models out of the box. You start paying when you pick a paid model — through OpenCode Go, OpenCode Zen, or your own provider key. Neither Go nor Zen is required.

OpenCode Go

Go is a model subscription: $10 a month, or $40 for Go Plus. It covers mostly open models (DeepSeek, Kimi, GLM, Qwen, MiniMax, MiMo), plus Grok, GPT Luna and Claude Haiku 5.5 — and the list changes as OpenCode tests new ones. You subscribe in the OpenCode console, copy the API key, and add it with /connect. Only one member of a workspace can subscribe.

The limits are dollars, not requests. Each model has a monthly allowance — from $15 to $80 on Go, depending on the model — and its usage is counted at that model's per-token price. Five hours may use 20% of the allowance and a week 50%. Go Plus raises the allowance on every model; the prices stay the same. OpenCode says models it could not get a discount on have the lower allowance.

When you reach a cap, the free models keep working, or you can turn on "Use balance" in the console and let Go spend your Zen credit. Go is designed for coding agents: OpenCode monitors traffic for abuse, and lists the other agents it has validated, Claude Code and Codex among them.

OpenCode Zen

Zen is a gateway to the models OpenCode has tested, billed per token from a prepaid balance — Claude, GPT, Gemini, Grok and the open models, some of them free for a limited time. OpenCode says it sells at cost and passes card fees through: 4.4% plus $0.30 a transaction. The first top-up is $20, and when the balance drops below $5 it reloads $20 unless you change or turn off auto-reload. You can set a monthly limit for the workspace and for each member, and use your own OpenAI or Anthropic key inside Zen, billed by the provider.

Zen lists Claude at Anthropic's rates — Opus 5.5 at $4 input and $20 output per million tokens. The same model is not always the same price on Zen and Go: DeepSeek V4 Pro is $1.74 per million input tokens on Zen and $0.66–1.32 on Go, the same range DeepSeek's own API lists, off-peak to peak.

Your own key, or a plan you already have

OpenCode connects to more than 75 providers with an API key, and runs local models. The provider bills you at its own price. OpenCode also lists subscriptions you can sign in with: ChatGPT Plus, GitHub Copilot and GitLab Duo.

Can I use Claude Pro or Max in OpenCode?

No. OpenCode's providers page says plugins that let OpenCode use a Claude Pro/Max plan exist, that "Anthropic explicitly prohibits this", and that OpenCode stopped bundling them in version 1.3.0. Claude Code's own documentation says third-party developers may not route requests through Free, Pro or Max credentials on behalf of their users. Claude in OpenCode is billed per token: an Anthropic API key, or Zen.

What an OpenCode call costs, measured

On my machine, OpenCode recorded 9,309 calls between 23 May and 18 June 2026, 84% of them on DeepSeek V4 Pro through DeepSeek's own API and 16% on MiniMax M3 through OpenRouter — my own keys, not Go or Zen. OpenCode computes each call's cost itself, and Caprock reads it from OpenCode's database. Read on 3 October 2026: the median call cost $0.002, the median active day $2, and all 21 active days $156, of which DeepSeek V4 Pro was $129. The busiest day came to $47.

The median call hides the long ones: the mean call cost $0.017. That is why this page has no "calls a day" calculator — multiplying by the median would understate a month several times over.

Would Go have covered it? As an estimate: priced at Go's DeepSeek V4 Pro rates — peak or off-peak by the hour each call ran — those calls come to about $94, not the $129 OpenCode recorded at the rates it applied on the day. That is more than six times Go's $15 monthly allowance for the model and more than Go Plus's $60, in under four weeks. The busiest seven days come to about $50 at those rates, over Go Plus's weekly $30, and the median day, about $3, already fills Go's five-hour window of $3.

For comparison, on the same machine the median Claude Code call cost $0.094 and the median Codex call $0.032, at Anthropic's and OpenAI's list prices. Different projects, months and models — DeepSeek V4 Pro is not Opus 5. The side-by-side is OpenCode vs Claude Code; all five agents are in AI coding agents compared.

See what your OpenCode sessions cost

Caprock finds OpenCode's database, opens it read-only and shows each session's cost next to Claude Code and Codex — per session, per repository, per day — with nothing to configure. caprock export writes every call:

$ caprock export events --format tsv --out events.tsv

OpenCode's figure is its own; Claude Code's and Codex's beside it are list-price estimates, as Claude Code's /usage is. Caprock records every session whether or not it started it. Free, open source; nothing leaves your machine. Install Caprock.