OpenCode Go on OpenClaw: API Key Setup, Models & Cost

What Is OpenCode Go?

OpenCode Go is opencode.ai’s subscription for open coding models: one API key and one monthly price for Kimi, DeepSeek, GLM, Qwen, MiniMax, MiMo and more, served from the opencode.ai/zen/go endpoint. That endpoint is why it is often called “Zen Go”; OpenCode Zen itself is the separate pay-per-token gateway. Go is built for coding agents, which is what OpenClaw is.

What It Costs

There are two separate things to pay for: the models, and the machine your agent runs on.

Saving your OpenCode Go key on OpenClaw Launch is free and needs no plan. What needs a plan is keeping an agent online after the 30-minute trial ends — a Telegram, Discord or WhatsApp bot that answers around the clock.

Models on OpenCode Go

The Go catalog changes often — opencode.ai retires and adds models every few weeks. These were live on 30 September 2026:

• Kimi K3 (kimi-k3) and Kimi K2.7 Code (kimi-k2.7-code) — Moonshot AI’s newest flagship and its coding-tuned sibling.
• DeepSeek V4 Pro (deepseek-v4-pro) and DeepSeek V4.1 Flash (deepseek-v4.1-flash) — the strongest DeepSeek model and a fast, cheap everyday one.
• GLM-5.3 (glm-5.3), GLM-5.2 and GLM-5.3 Flash — Zhipu AI’s GLM line, strong at agentic coding and bilingual work.
• Qwen3.8 Max (qwen3.8-max), Qwen3.7 Plus (qwen3.7-plus) and Qwen3.8 Flash — Alibaba’s Qwen family from top tier to fast.
• MiniMax M3 (minimax-m3) — MiniMax’s long-context flagship.
• MiMo V2.6 Pro / Flash — Xiaomi’s MiMo reasoning models.
• GPT 6 Luna, Grok 4.7 and Grok 4.6 — these use the OpenAI Responses API. OpenClaw’s built-in provider already knows that, so no extra setting is needed.

Kimi K2.6, Qwen3.6 Plus and GLM-5.1 have been retired from Go and now return an error, even though some older pages still list them. On the $10 Go plan the premium models (Kimi K3, DeepSeek V4 Pro, GLM-5.3, Qwen3.8 Max, GPT and Grok) each have a smaller monthly allowance than the flash models, so a flash model is the better default for an always-on bot. Current limits are on opencode.ai/docs/go.

How to Get an OpenCode Go API Key

1. Open opencode.ai/go — opencode.ai/go, and sign in or create an account.
2. Subscribe to Go or Go Plus — the key only reaches the Go models while the subscription is active.
3. Add a service account — in the OpenCode console, open the Keys tab and click Add Service Account. Any name works, for example hermes. Then open the service account you just created.
4. Create a key — click Add API Key, give it any name, leave Expiration on Never and Permissions on the default All, then click Create. Copy the key straight away and keep it private. New keys start with oc_sk_.

Option 1: Add the Key on OpenClaw Launch

1. Sign in and open the API Keys page.
2. Find the OpenCode Go card, paste your key and click Save. The box shows sk-... as a placeholder, but an oc_sk_ key pastes in exactly the same way. The key is encrypted and saved, then checked against OpenCode Go; the card tells you if that check failed.
3. Click “Set as primary model” and choose a model. The card lists only models OpenCode Go is serving right now, newest first.
4. If you have no agent yet, deploy one from the home page. Running agents switch to the new model without a redeploy.

The steps are the same for OpenClaw and Hermes Agent instances: the key is saved once per account and applies to both.

Option 2: Self-Host with the Built-In Provider

OpenClaw ships an opencode-go provider, so a self-hosted install needs only the key and a model — no custom provider block. The key goes in OPENCODE_API_KEY, the same variable OpenClaw uses for OpenCode Zen.

# Store the key and pick a model
openclaw onboard --opencode-go-api-key "$OPENCODE_API_KEY"
openclaw config set agents.defaults.model.primary "opencode-go/kimi-k3"

# Check which Go models your install can see
openclaw models list --provider opencode-go

Or set the same thing directly in openclaw.json:

{
  env: { vars: { OPENCODE_API_KEY: "oc_sk_..." } },
  agents: { defaults: { model: { primary: "opencode-go/kimi-k3" } } }
}

Use OpenClaw 2026.9.4 or newer. OpenCode Go rejects requests that carry no session header, and 2026.9.4 is the first release that sends one on every call.

On self-hosted Hermes Agent the provider and key have different names: OPENCODE_GO_API_KEY in ~/.hermes/.env, then model.provider: opencode-go and model.default: opencode-go/kimi-k3 in ~/.hermes/config.yaml, on Hermes v2026.9.7 or newer. The Hermes Agent guide has the full commands.

Troubleshooting

401 Unauthorized / Invalid Key

• Confirm you copied the whole key (new keys start with oc_sk_) with no trailing spaces or newlines.
• Check that your Go subscription is active on opencode.ai. A key without an active Go plan cannot reach the Go models.
• Create a fresh key under Keys in the OpenCode console and save the new value.

400 MissingSessionID

• Self-hosted only: your OpenClaw is older than 2026.9.4. Update it; managed instances on OpenClaw Launch already handle this.

Model Not Found or Model Retired

• The model has probably been retired from Go. Kimi K2.6, Qwen3.6 Plus and GLM-5.1 all were. Pick a current one from the API Keys card, or run openclaw models list --provider opencode-go when self-hosting.
• Self-hosted model refs need the provider prefix: opencode-go/deepseek-v4-pro, not deepseek-v4-pro.

Quota or Rate Limit Errors

• Go meters usage in dollars per model over a 5-hour, weekly and monthly window. Premium models run out first on the $10 plan — switch to a flash model or upgrade to Go Plus.
• Check usage on your opencode.ai dashboard.

Frequently Asked Questions

Do I need to pay OpenClaw Launch if I already have OpenCode Go?

Only to keep an agent running on our servers. OpenCode Go pays for the models; OpenClaw Launch pays for the always-on machine. You can try an agent free for 30 minutes, then Lite is $3 for the first month and $6/month after. Or self-host OpenClaw with the built-in opencode-go provider and pay us nothing.

Does this work on Hermes Agent?

Yes. On OpenClaw Launch the same saved key applies to Hermes Agent instances, with identical steps. Self-hosted Hermes uses its own opencode-go provider with the key in OPENCODE_GO_API_KEY; see the Hermes Agent OpenCode Go guide.

Which OpenCode Go model is best for OpenClaw?

For an always-on bot, a flash model such as deepseek-v4.1-flash or glm-5.3-flash stretches the $10 plan furthest. For hard coding work, kimi-k3, deepseek-v4-pro and glm-5.3 are the strongest, but they draw down a smaller allowance. The AI coding plans compared guide sets Go against the alternatives.

Can I use OpenCode Go alongside other provider keys?

Yes. Each provider key is stored separately, so an OpenCode Go key can sit next to OpenRouter, Groq or a ChatGPT subscription. Switch with “Set as primary model” or the instance model selector.

What's Next?

• OpenCode Go on Hermes Agent — the same plan, set up for Hermes Agent
• AI coding plans compared — OpenCode Go against the other coding subscriptions
• OpenClaw BYOK overview — every provider you can bring your own key for
• Pricing — free trial, Lite and Pro