OpenCode Go explained: the $10 open-model subscription, how its limits work, and using it in Claude Code

OpenCode Go is OpenCode's own model subscription: a fixed monthly price for a set of mostly open coding models such as GLM, Kimi, Qwen, DeepSeek and MiniMax, used inside OpenCode. It is optional. OpenCode itself is free and open source (MIT licence) and works without Go, with your own API keys or a third-party service; see the end of this article.

Prices, models and limits below come from OpenCode's official documentation, checked on 2026-11-14. OpenCode says models and limits change over time, so check the Go docs before you subscribe.

The three plans

Go and Go Plus are only for workspaces with one member, such as the Personal workspace created when you sign up. You can upgrade to Go Teams at any time, and the unused part of the month's allowance is credited.

Which models it includes

The official list has 32 models, in roughly three groups:

• Open or open-weight models: GLM-5.3, GLM-5.3-Flash, GLM-5.2, Kimi K3.1 Pro, Kimi K3, Kimi K2.7 Code, Qwen3.8 Max, Qwen3.8 Flash, Qwen3.7 Plus, DeepSeek V4 Pro, DeepSeek V4.1 Flash, MiniMax M3, MiniMax M3.1 Flash, MiMo, LongCat, the Hy series, Step 5 Preview and others.
• A few small proprietary models: Claude Haiku 5.5, GPT 6 Luna, GPT 5.6 Luna, Gemini 4 Argon, Grok 4.7.
• Free for a limited time: Beam Free, Iceman Contributor Free, LongCat 2.5 Free.

Go doesn't include the providers' flagship models. For Claude it has only Haiku 5.5; for GPT, only the two Luna models. Anything else means paying per token from a Console balance, or using another service.

How usage is counted

This is the part people misread most: Go limits are measured in dollars, not requests.

• Each model has a monthly limit in dollars and a price per million tokens. Your usage is priced at those rates and counted against the limit.
• Each model also has two shorter caps: at most 20% of the monthly limit in any 5 hours, and 50% in a week. You can't spend a month's allowance in a day or two.
• Every response includes an opencode.allowance field showing how much that request counted. The docs don't spell out exactly how usage across different models combines, so go by the remaining allowance shown in the Console.

Monthly limits on the Go ($10) plan for some common models, with OpenCode's estimated request counts:

Things to note:

• The stronger the model, the fewer requests you get. Kimi K3 allows about 110 requests per 5 hours on Go, and one involved task in a coding agent can use a good share of that. Cheap Flash-class models allow far more.
• The estimates assume most of each request hits the cache. OpenCode assumes 50k to 90k cached tokens per request plus a few hundred to about a thousand new input tokens. With fewer cache hits or longer contexts you get fewer requests.
• Some prices are temporary. Gemini 4 Argon is half price until 2026-11-30. The two DeepSeek models have peak and off-peak prices; peak is 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday.
• Go Plus and Go Teams use the same token prices; only the limits are higher.

When you run out

• Switch to a free model. OpenCode's free models have a daily request limit but don't use your Go allowance. They only work inside OpenCode; other tools get HTTP 403.
• Turn on Extra Usage. On the Go page in the Console. Once your limits are used up, requests are charged to your balance at the listed prices instead of being blocked. Go Teams has it on by default.
• Upgrade to Go Plus.

Connecting it

In OpenCode:

1. Sign in to the OpenCode Console, subscribe to a plan and copy your API key.
2. In OpenCode, run /connect, choose OpenCode Go and paste the key.
3. Run /models and pick a model.

To set the model in a config file, use opencode-go/model-id:

{
  "$schema": "https://opencode.ai/config.json",
  "model": "opencode-go/kimi-k3"
}

In Claude Code: Go works with Claude Code too. Create a personal API key in the Console (it starts with oc_sk_) and add this to ~/.claude/settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://opencode.ai/inference/go/anthropic",
    "ANTHROPIC_AUTH_TOKEN": "oc_sk_your_key",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "minimax-m3",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-haiku-5-5"
  }
}

Use ANTHROPIC_AUTH_TOKEN, not ANTHROPIC_API_KEY. If Claude Code is signed in to a Claude account, run /logout first so it doesn't send that login. Then run /status to confirm the URL and key are active.

Other tools: OpenCode has also validated Codex, Hermes and Pi; Kilo Code CLI needs a recent build. Go requires the client to send a stable session ID for each conversation, and tools that can't can't use Go models, only pay-as-you-go ones. OpenCode lists GitHub Copilot Chat and Kimi Code among the clients with known problems.

Payment, regions and refunds

• Payment: the Console bills through Stripe and accepts cards, Link, Alipay, UPI and WeChat Pay. You can also top up a balance and pay for Go from it, but top-ups carry a 4.6% + $0.31 processing fee, so adding $50 costs $52.61.
• Regions: OpenCode says Go is designed primarily for international users. By default a workspace only allows Europe and United States endpoints. Workspaces that have never saved privacy settings also allow Global endpoints for requests from China. The two DeepSeek models need Global selected under Regions in Settings › Privacy. Beam Free isn't available in mainland China, Hong Kong or Macau, among other places.
• Privacy: paid endpoints that train on your requests are blocked by default. The Muse Spark models and Iceman Contributor Free use your requests for training and must be turned on by hand. Requests to Claude Haiku 5.5, the GPT Luna models and Grok 4.7 are retained for 30 days; most other models retain nothing.
• Cancelling: cancel any time; the plan stays active until the end of the period you paid for.
• Refunds: you may qualify if you ask within 14 days of the charge and have used almost none of that period's allowance.

Is it worth it?

A good fit if you:

• code mostly in OpenCode, and GLM, Kimi, Qwen, DeepSeek or MiniMax models are enough for you;
• want a fixed monthly cost instead of tracking tokens;
• use a lot: Go Plus at $40 gives more allowance per dollar.

Less of a fit if you:

• mainly want the providers' flagship models, which Go doesn't include;
• often run long tasks in one go, where the 5-hour (20%) and weekly (50%) caps get in the way;
• use a tool that doesn't send a session ID, which can't use Go models.

Using OpenCode without Go

OpenCode works with many model services. Enter your own Anthropic, OpenAI or other keys in /connect, or add a custom provider in opencode.json for any Anthropic- or OpenAI-compatible third-party API. You can set up Go and other services side by side and switch between them in /models.

To use CCSub, follow the OpenCode setup guide: add a provider in opencode.json with baseURL set to https://www.ccsub.net/api/v1. It is billed per token, with no 5-hour or weekly caps. For Claude Code, see the Claude Code setup guide.