OpenCode Go: price, models and usage limits

How much does OpenCode Go cost?

OpenCode Go pricing has two plans, both monthly, listed on opencode.ai/go:

You can cancel at any time. OpenCode says you may get a refund if you were charged in the last 14 days and haven't used any of that period's allowance. Go and Go Plus are for workspaces with one member, such as the personal workspace you get when you sign up, and the Go docs say a workspace with a Go subscription can't invite members.

Go is optional. OpenCode itself is free and open source, and it works with other providers too. Go is the subscription the OpenCode team sells for its own tested model lineup, and the docs say it's designed mainly for international users. The page lists prices in USD only.

Which models does OpenCode Go include?

OpenCode Go includes 30 models. Most come from open-weight families such as GLM, Kimi, DeepSeek and MiniMax, and a few are closed models: Claude Haiku 5.5, GPT 6 Luna, GPT 5.6 Luna, Grok 4.7 and Gemini 4 Argon. Here are some of the OpenCode Go models with their monthly limits and OpenCode's own estimate of requests per 5 hours, from the Go docs:

The rest of the list is GLM-5.2, Kimi K2.7 Code, LongCat-2.0, Step 5 Preview, MiMo-V2.6-Flash, MiMo-V2.6-Pro, MiMo-V2.5, MiMo-V2.5-Pro, MiniMax M3, Muse Spark 1.3 Contributor, Muse Spark 1.2 Contributor, Qwen3.8 Flash, Qwen3.7 Plus, GPT 5.6 Luna, Hy4 preview and Hy3, plus two free promotional models, Beam Free and Iceman Contributor Free. OpenCode says the list changes as it tests and adds models.

The request estimates assume a typical agent request with a large cached prompt, so treat them as a rough guide. The larger models, like Kimi K3 at 110 and GLM-5.3 at 220, give you a few hundred requests or fewer per 5 hours. The Flash models give you thousands.

What are the OpenCode Go limits?

OpenCode Go limits are monthly dollar amounts per model, and you can't spend them all at once. The docs set three windows for each model: 20% of its monthly limit per 5 hours, 50% per week, and 100% per month. Requests are counted at each model's token price, which is the same on Go and Go Plus, and every response reports how much it counted as opencode.allowance.

Take GLM-5.3-Flash. It has a $60 monthly limit on Go, so you can use $12 of it in any 5 hours and $30 in a week. Kimi K3's limit is $15, so its 5-hour window is $3.

Some limits are low because of what OpenCode pays. Its docs say most models get high allowances through bulk discounts and reserved GPU capacity. For new models, or ones whose public price is already discounted, you get "a little more than if you paid the model providers directly", which is why those allowances are smaller.

When you hit a limit, you can keep using the free models, or turn on Extra Usage, which falls back to a pay-as-you-go balance in OpenCode's Console instead of blocking requests.

Can you use OpenCode Go with Claude Code or other agents?

Yes. Go is an API key, so it works in OpenCode with /connect and in other agents that send normal coding-agent traffic. The docs list Claude Code, Codex, Hermes, Pi, ZCode, jcode and Kilo Code CLI as validated clients. For Claude Code, you point ANTHROPIC_BASE_URL at OpenCode's Go endpoint and use your Go key as the auth token.

OpenCode monitors traffic for abuse. It asks clients to identify themselves with their own user agent and send a stable session ID, which helps it route requests and cache prompts.

What about privacy on OpenCode Go?

By default, Go blocks paid endpoints that train on your requests and allows only Europe and United States regions. To use DeepSeek V4.1 Flash or V4 Pro, a workspace owner selects the Global region in privacy settings. The two Muse Spark Contributor models are discounted in exchange for letting Meta train on your prompts, so they need a separate opt-in. The Go page lists each model's training and retention terms.

Where Buildful fits

OpenCode Go is a model plan: you get the models, and you drive them from your own agent on your own machine. Buildful is a background coding agent. You describe a task, and Buildful runs it in a private cloud sandbox, runs your checks, tests UI changes in a real browser, and opens a pull request on your GitHub repo.

Buildful Pro is $10 a month for 1,000 tasks, or ₹750 in India, and Team is $40 a month for 3,000 tasks shared by unlimited members (₹3,500 in India). Smart Mode picks the right open-weight model for each task, and you can pick a frontier model like Claude yourself. Four of the models Go carries, Kimi K3, DeepSeek V4.1 Flash, GLM 5.3 Flash and MiniMax M3.1 Flash, are also in the Buildful Pass.

On price, Go is the lower way in. Buildful is for when you'd rather hand off the task than run the agent. If you're choosing an agent to use Go with, OpenCode vs Claude Code compares the two most common ones.

Frequently asked questions

Is OpenCode Go free?

No. OpenCode Go costs $10 a month, or $40 for Go Plus. OpenCode itself is free and comes with a few free models, such as Big Pickle, that you can use without Go.

What's the difference between OpenCode Go and Zen?

Zen is OpenCode's pay-as-you-go model service, where you pay for what you use. Go is a $10 monthly subscription with set allowances on a curated list of models.

Does OpenCode Go include Claude?

Only Claude Haiku 5.5. Go doesn't include Claude Sonnet, Opus or Fable. For those, you'd use an Anthropic API key or a Claude plan.

Can I share OpenCode Go with my team?

No. Go and Go Plus are for single-member workspaces, and a workspace with a Go subscription can't invite members.

What happens when I reach my OpenCode Go limit?

Requests to that model stop until its window resets, unless you turn on Extra Usage, which bills your pay-as-you-go balance instead. You can also switch to the free models in the meantime.

Can I get a refund on OpenCode Go?

OpenCode says you may qualify for a refund if you were charged within the last 14 days and haven't used your Go allowance in that billing period. You request it by contacting OpenCode.