Short answer: OpenCode is an open source (MIT) terminal client maintained by Anomaly, with 195,000 GitHub stars: you choose the AI provider (Anthropic, OpenAI, Google, local via Ollama, or any OpenAI-compatible endpoint) and pay for it separately, either with your own key, through the Go plan (US$10/month, curated open models), or through Zen (pay-as-you-go, includes frontier models). Verboo Code is the provider: dedicated GPU, open models like GLM-5.2 and Kimi K2.7, unlimited tokens on a flat subscription. You don't have to pick a side: you can keep the OpenCode client you already use and just swap the engine behind it for Verboo Code. If you want the highest possible benchmark ceiling on a single hard task, stick with Zen or your own key running a frontier model. If you run long agent sessions and want to stop watching a meter, point OpenCode at Verboo Code.
OpenCode and Verboo Code side by side
| Attribute | OpenCode | Verboo Code |
|---|---|---|
| What it is | Open source terminal client, you bring the provider | Its own provider with a terminal CLI, dedicated GPU |
| License | Open source (MIT), maintained by Anomaly | Closed, SaaS subscription |
| Billing model | Bring your own key, or Go US$10/month, or Zen pay-as-you-go | Flat subscription in Brazilian reais |
| Entry price | Free client; Go starting at US$10/month | Junior R$99/month |
| Model catalog | Any provider: Anthropic, OpenAI, Google, local, custom endpoint | 11 open models (GLM-5.2, Kimi K2.7, DeepSeek V4 and others) |
| Token limit | Depends on the provider and plan you pick | Unlimited on every plan |
| Context | Depends on the chosen model | 1M tokens on almost all of them (exception: qwen3.8-27b, 262k) |
| OpenAI-compatible endpoint | Accepts a custom provider natively, via opencode.json | Exposes its own OpenAI-compatible router, ready for any client to point at |
When OpenCode is the right choice
- You want to choose between any AI provider (Anthropic, OpenAI, Google, local via Ollama) and switch whenever you want.
- You already pay for a frontier model's API and want to use it inside a terminal client, without a closed catalog.
- You prefer a fully open source client, without depending on a closed company behind the tool.
- You're fine managing API cost on your own, either with a direct key or the dollar cap on the Go plan.
When Verboo Code is the right choice
- You run long agent sessions, big refactors or lots of files in context, and don't want to watch a token bill mid-task.
- You prefer a flat subscription over managing a prepaid dollar balance.
- You want to keep the client you already use, including OpenCode itself, and just swap the engine behind it.
- You don't need the market's most expensive model for most day-to-day tasks, you just want tokens that don't run out.
Where each one actually wins
OpenCode has no model of its own: it's a client that talks to any provider, and that's exactly where its biggest raw benchmark advantage lives. Running through Zen (pay-as-you-go) or your own key, you can plug in Claude Opus 4.8 or GPT-5.6, the strongest models on the market today. On Terminal-Bench 2.1, Opus 4.8 scores 85.0 points, against 81.0 for GLM-5.2, the best open model available today on Verboo Code, according to Z.ai's official GLM-5.2 launch on June 13, 2026. For anyone who wants the highest possible ceiling on a single hard task, the path is OpenCode with a frontier model plugged in.
But that flips on a different benchmark. On SWE-bench Pro, the same GLM-5.2 beats GPT-5.5: 62.1 against 58.6 points, in the same Z.ai launch (06/13/2026). And the benchmark scoreboard doesn't show the bill: running Opus 4.8 or GPT-5.6 through Zen or your own key is billed per token, in dollars, with no cap, and an agent session with dozens of files in context and a long history burns through that fast. OpenCode's Go plan (US$10/month) solves that with a usage cap per period, but it only covers the curated open model catalog, not frontier models, which sit outside Go and only come in through Zen.
Here there's no mandatory side. OpenCode natively accepts a custom provider: a block in opencode.json using the @ai-sdk/openai-compatible package, pointing baseURL at any endpoint that speaks the OpenAI protocol, per OpenCode's own official documentation. Verboo Code exposes exactly that endpoint, at code.verboo.ai/router/v1, authenticated with a vbk_ key generated on its own dashboard. Once configured, OpenCode stays OpenCode, TUI and commands included, only the model behind it starts running on dedicated GPU with unlimited tokens.
The choice, then, isn't "which tool thinks better," it's "how many times can you retry without recalculating your budget." For a rare, hard ticket, where 4 points on Terminal-Bench decide the outcome, Opus 4.8 through Zen still wins. For a recurring agent session, where the cost of getting it wrong and trying again is what stalls the work, that same OpenCode pointed at Verboo Code trades a frontier model for an unlimited ceiling at a flat price.
How to point OpenCode at Verboo Code's engine
1. On the code.verboo.ai dashboard, generate an API key (prefix vbk_)
2. Export the key as an environment variable:
export VERBOO_API_KEY="vbk_your_key_here"
3. Create (or edit) opencode.json at the project root,
or at ~/.config/opencode/opencode.json:
{
"$schema": "https://opencode.ai/config.json",
"provider": {
"verboo-code": {
"npm": "@ai-sdk/openai-compatible",
"name": "Verboo Code",
"options": {
"baseURL": "https://code.verboo.ai/router/v1",
"apiKey": "{env:VERBOO_API_KEY}"
},
"models": {
"glm-5.2": { "name": "GLM-5.2" },
"kimi-k2.7": { "name": "Kimi K2.7" }
}
}
}
}
4. Run "opencode" and switch to the verboo-code provider in the model picker
5. To check which models your key unlocks:
curl https://code.verboo.ai/router/v1/models \
-H "Authorization: Bearer $VERBOO_API_KEY"
Everything else about OpenCode stays the same: TUI, plan-then-build, MCP and LSP integration. Only the model behind the responses moves to Verboo Code's dedicated GPU, with no per-session token counter.
Frequently asked questions
Can I use OpenCode with Verboo Code's models?
Yes. Set up a custom provider in opencode.json with @ai-sdk/openai-compatible, pointing baseURL at https://code.verboo.ai/router/v1 and a vbk_ key generated at code.verboo.ai. The rest of OpenCode stays the same.
Is OpenCode free?
The client is, under the MIT license. What costs money is the model behind it: your own key for any provider, the Go plan (US$10/month, curated open models), or Zen (pay-as-you-go, includes frontier models like Claude and GPT).
What's the difference between OpenCode's Go plan and Verboo Code?
Go charges US$10/month with a usage cap per period over a catalog of open models. Verboo Code charges a flat subscription in Brazilian reais, starting at R$99/month, with unlimited tokens inside the plan and no usage cap per period.
Do I need to switch tools to get unlimited tokens?
No. You can keep OpenCode and point a custom provider at Verboo Code, or use Verboo Code's own CLI (verboo) if you'd rather have an agent that's already pre-configured, without setting up the provider by hand.
If the math that decides isn't which model thinks better, but how many times you can retry without blowing your budget, Verboo Code solves that with unlimited tokens at a flat price, either through its own terminal or by pointing OpenCode at its engine.
