OpenCode pricing: free, Go at $10, Zen, and the plans it takes
OpenCode is free; you pay for the model: Go and Go Plus, pay-as-you-go Zen, ChatGPT, Copilot or an API key, and why a Claude Pro or Max plan doesn't work.
OpenCode itself is free and open source (MIT), and comes with a few free models capped at 200 requests a day. What you pay for is the model. As of October 9, 2026, that is one of: OpenCode’s own Go subscription at $10 a month (or Go Plus at $40) for a set of about thirty coding models; its pay-as-you-go catalogue, called Zen on the main docs and OpenCode Console on the version 2 docs, at the providers’ own token prices plus a card fee; or a plan or key you already have.
Plans you can bring: ChatGPT Plus or Pro (OpenAI lists OpenCode among the apps that can use your plan), GitHub Copilot, GitLab Duo, a SuperGrok subscription, and an API key from almost any provider, OpenRouter included. A Claude Pro or Max plan is the exception: OpenCode has had no Claude sign-in since version 1.3.0 (March 2026), Anthropic’s terms bar other developers from offering one, and Claude models go in by API key, although OpenCode’s home page still says otherwise.
What is free
The program is free under the MIT licence, with no account needed. The home page’s answer to what it costs: “OpenCode is 100% free to use. It also comes with a set of free models. There might be additional costs if you connect any other provider.” The free models change: on October 9 they were Big Pickle and a dozen “free for a limited time” previews (Nemotron, MiMo, Ling, LongCat and others), and the Go page puts their quota at “200 requests/day”. Two catches sit in the docs’ privacy notes: on several of them, “During its free period, collected data may be used to improve the model,” and the free Nemotron endpoints say “Trial use only — do not submit personal or confidential data.” Free models are fine for trying OpenCode out; for work you would not paste into a public form, pay for a model.
OpenCode Go and Go Plus
Go is a monthly subscription to a curated lineup of coding models, sold mostly as open models; the list also includes Claude Haiku 5.5, GPT 6 Luna, GPT 5.6 Luna and two Grok models.
| Go | Go Plus | |
|---|---|---|
| Price | $10 a month | $40 a month |
| Models | The same lineup: 32 in the docs (the Go page says 31), among them Kimi K3, GLM-5.3, DeepSeek V4.1, Qwen3.8, MiniMax M3, Grok 4.7 and Claude Haiku 5.5 | |
| How limits work | A monthly dollar amount per model, with 20% of it usable in any 5 hours and 50% in a week; token prices are the same on both plans | |
| Monthly limit, Kimi K3 | $15 (about 110 requests per 5 hours, by OpenCode’s estimate) | $60 (about 440) |
| Monthly limit, DeepSeek V4.1 Flash | $60 (about 26,000 requests per 5 hours) | $120 (about 52,000) |
| At the limit | Fall back to the free models, or switch on paying from your pay-as-you-go balance | |
A few terms worth knowing before you subscribe. Go is for one person: “A workspace with a Go subscription cannot invite members.” It isn’t tied to OpenCode: “Yes, you can use Go with any agent,” and the docs list Claude Code, Codex, Hermes and Pi among the clients OpenCode has checked, with no promise they keep working. You can cancel any time, and a refund is possible within 14 days if you haven’t used that period’s allowance. Neither the Go page nor its docs offer a free trial. The docs add that the service “is designed primarily for international users”, and its models are hosted in the US.
Pay as you go: Zen, or OpenCode Console
The second offer is a catalogue of models OpenCode has tested, paid per token from a prepaid balance. The main docs call it OpenCode Zen; the version 2 docs call the same catalogue OpenCode Console and never use the word Zen. OpenCode says it adds no markup (“Zen is not for profit”) and passes on the card fee, which its pages state two ways: 4.4% plus $0.30 a transaction on the models pages, 4.6% plus $0.31 on the billing page and the Zen page ($1.23 on a $20 top-up). You add credit (the Zen page starts at $20; the version 2 billing page offers $10, $50, $100 or an amount you choose), and by default it tops up another $20 when the balance drops under $5; you can change or switch that off, set monthly limits, and requests fail once the balance hits zero. A few prices per million tokens, input and output, on October 9: Claude Opus 5.5 $4 and $20, Claude Sonnet 5.5 $2 and $10, GPT 6.1 Sol $2 and $10, Gemini, Grok, Qwen, DeepSeek and the rest beside them.
Where the models run is stated two ways as well: “All Zen models are hosted in the US” on the Zen page, “hosted in the US and EU” in the docs. Both say the providers keep no data and don’t train on it, with listed exceptions: requests to OpenAI’s and Anthropic’s APIs are kept for 30 days.
Two other names turn up on OpenCode’s site. OpenCode Black, a subscription to the frontier models, showed only “Black plan enrollment is temporarily paused” and no price on October 9. Enterprise is per seat, with no charge for tokens that go through a company’s own model gateway.
Plans and keys you already have
| What you have | Works in OpenCode? | How it connects |
|---|---|---|
| ChatGPT Plus or Pro | Yes; usage counts against your plan’s limits | Sign in with ChatGPT (OpenAI, then the ChatGPT option) |
| GitHub Copilot | Yes; some models need Pro+ | Device sign-in at github.com/login/device |
| GitLab Duo | Yes, experimental; Premium or Ultimate | GitLab sign-in or a personal access token |
| SuperGrok or an X plan with Grok API access | Yes | Device sign-in |
| An API key (Anthropic, OpenAI, Google, OpenRouter and 75+ more) | Yes, billed by the provider | opencode auth login, or the provider’s environment variable |
| A local model (Ollama, LM Studio, vLLM) | Yes, free | Version 2 finds a running Ollama, LM Studio or vLLM by itself |
| Claude Pro or Max | No | Anthropic’s models go in by API key only |
On the ChatGPT row, OpenAI’s Sign in with ChatGPT page names OpenCode under its open-source integrations and says “eligible ChatGPT Plus and Pro subscribers can also choose to use their ChatGPT plan for AI requests.” The usage comes out of the same allowance Codex draws on: “Connecting an app does not add a new allowance.” You can cap an app at a share of your weekly usage, and disconnect it under ChatGPT’s Security and login settings. In OpenCode 2.0.26 the sign-in is listed as “Sign in with ChatGPT”, beside two older “Codex (legacy)” methods; the version 2 docs mention only the legacy one. OpenAI’s pages name Plus and Pro; the ChatGPT plans guide has what each plan includes for Codex.
Why a Claude plan doesn’t work
OpenCode used to ship a plugin that signed in with a Claude account. Version 1.3.0 (March 22, 2026) removed it, in a change titled “anthropic legal requests”, and OpenCode’s providers page now says: “There are plugins that allow you to use your Claude Pro/Max models with OpenCode. Anthropic explicitly prohibits this.” Version 2.0.26 offers Anthropic by API key or ANTHROPIC_API_KEY and nothing else. Anthropic’s side is its legal and compliance page: “Anthropic does not permit third-party developers to offer Claude.ai login into their own applications, or to route requests through Free, Pro, or Max plan credentials on behalf of their users.”
Two pages blur that. OpenCode’s own home page still answers “Can I use my existing AI subscriptions?” with “You can use your Claude Pro/Max, ChatGPT Plus/Pro, or GitHub Copilot subscriptions.” And Anthropic’s help article on logging in calls an API key “the preferred way” for other tools while saying Anthropic “may at its discretion allow paid subscribers who have enabled usage credits to use certain third-party tools,” drawn “from usage credits rather than subscription limits”; it names no tool, and OpenCode offers no such sign-in. A third Anthropic page, on the Agent SDK and your plan (updated October 8, 2026), says “You can still use the Claude Agent SDK, claude -p, and third-party apps with your subscription limits,” without saying which apps it means. The way to use Claude models in OpenCode is an Anthropic API key, billed per token. If you have Max, one thing helps: since October 7, 2026, Max plans include monthly credits for the Claude API, $100 on Max 5x and $200 on Max 20x, and “Everyone with an API key in the linked organization draws from the same balance.” If you want Claude on your Claude plan, Claude Code is the tool that plan covers.
Connecting one
Inside OpenCode, /connect lists the providers and their sign-in methods, and /models picks a model. From a terminal, version 2 does the same with opencode auth login, for example opencode auth login anthropic --method key, and keeps keys and tokens in its local database; version 1 stores them in ~/.local/share/opencode/auth.json. For Go or the pay-as-you-go catalogue you sign in to OpenCode’s console, subscribe or add credit, copy the API key and choose OpenCode Go or OpenCode Console in /connect. On a server with no browser, the ChatGPT, Copilot and Grok sign-ins use device codes, which you finish on your phone; the remote server guide covers reaching OpenCode there once it is signed in.
Two versions, two sets of docs
OpenCode has two release lines. The home page’s install command gives you version 2 (@opencode/cli, 2.0.26 on October 8), whose docs live at opencode.ai/v2/docs; the older line (opencode-ai, 1.18.35) keeps the docs at opencode.ai/docs. opencode --version tells you which you have. For pricing the difference is mostly names (Zen against Console), but the pages also differ on details, which are listed above where they matter: the card fee, where Zen’s models are hosted, whether Go has 31 or 32 models, and what a Claude plan does.