{"repo": "bitdoze/bitdoze.com", "files": {"src/content/posts/opencode-go-plan.mdx": {"3d455461c96abd9d88dfb2ca79bc3ccde5f8e0b0": {"t": 1785595265.909729, "n": 11706, "h": "5d258b1691cf5d82", "c": "---\ndate: 2026-07-24T00:00:00Z\ntitle: \"OpenCode Go Review 2026: 16 AI Models for $10/Month (Grok 4.5, Kimi K3)\"\ndescription: \"OpenCode Go review after 3 weeks of daily use. One API key, 16 models including Grok 4.5, Kimi K3, and DeepSeek V4 Pro. Works with Hermes, OpenClaw, Pi Agent. $5 first month.\"\nimage: \"../../assets/images/26/05/opencode-go-plan.webp\"\ncategories: [\"ai\"]\nauthors: [\"Dragos\"]\ntags: [\"ai-tools\", \"opencode\", \"llm\"]\ncanonical: \"https://www.bitdoze.com/opencode-go-plan/\"\n---\n\nimport Button from \"@components/widgets/Button.astro\";\nimport Notice from \"@components/widgets/Notice.astro\";\nimport ListCheck from \"@components/widgets/ListCheck.astro\";\nimport Accordion from \"@components/widgets/Accordion.astro\";\nimport Tabs from \"@components/widgets/Tabs.astro\";\nimport Tab from \"@components/widgets/Tab.astro\";\nimport YouTubeEmbed from \"@components/widgets/YouTubeEmbed.astro\";\n\nI have been juggling API keys for coding agents for months. DeepSeek here, MiniMax there, OpenRouter balance to top up, three billing pages I forget which password goes to. It works. It is still annoying.\n\nOpenCode Go is basically one $10/month key ($5 the first month) for 16 models: Grok 4.5, Kimi K3, GLM-5.2, DeepSeek V4, MiniMax M3, and a pile of cheaper options. It plugs into OpenCode, Hermes, OpenClaw, Pi, or anything that speaks OpenAI-compatible APIs.\n\nI almost skipped it. Another subscription. Another thing to cancel later. At $10 I figured I would burn a week and move on if it sucked. Three weeks later I am still on it. Here is the honest version.\n\n\n\n## What OpenCode Go is\n\n\n\nOpenCode Go is a flat subscription from the OpenCode team: one API key, 16 models, $10/month after a $5 first month. No per-token math. No separate accounts at every model provider.\n\nModels are served from the US, EU, and Singapore. Providers claim zero retention, so your code is not supposed to train anyone's next model. That matters if you ship proprietary work.\n\nYou do not need OpenCode the IDE to use Go. Point any OpenAI-compatible client at the endpoint: [Hermes Agent](/hermes-agent-setup-guide/), [OpenClaw](/clawdbot-setup-guide/), [Pi Agent](/pi-coding-agent-setup-guide/), [Mastra](/build-ai-agent-mastra/), Agno, Codex, whatever you already run.\n\n\n\n## Which models are included\n\nThe current list (16 models):\n\n| Model | Notes |\n|-------|-------|\n| **Grok 4.5** | Strong on hard tasks, but the quota burns fast (use sparingly) |\n| **Kimi K3** | New Moonshot flagship; good at code, few requests per cap |\n| **GLM-5.2** | Zhipu's current flagship for coding agents |\n| GLM-5.1 | Previous GLM, same usage tier as 5.2 |\n| Kimi K2.7 Code | Coding-tuned Kimi with way more requests than K3 |\n| Kimi K2.6 | Fine daily driver if you do not need K3 |\n| MiMo-V2.5-Pro | Stronger Xiaomi coding model |\n| MiMo-V2.5 | Cheap and high request count |\n| MiniMax M3 | Solid agentic workhorse |\n| MiniMax M2.7 | Still my default for day-to-day agent work |\n| Qwen3.7 Max | Top Qwen tier from Alibaba |\n| Qwen3.7 Plus | Middle Qwen 3.7 option |\n| Qwen3.6 Plus | Older Qwen, still usable |\n| DeepSeek V4 Pro | Strong general coding, long context |\n| DeepSeek V4 Flash | Fast and cheap for quick jobs |\n| Hy3 | High request count open coding model |\n\n\nBoth are on Go now, but they only get about $15 of included usage per month instead of $60 like most of the roster. You will feel that on the 5-hour cap. I keep them for the hard problems and burn MiniMax, DeepSeek Flash, or MiMo when I am grinding through volume. Run `/models` in OpenCode if the list has shifted again.\n\n\nOpenCode does not dump every new model onto Go. They test for coding-agent use first, then add what holds up. The list will keep moving.\n\n## Usage limits\n\nGo is not unlimited. It uses dollar-value limits, not request counts. Your actual number of requests depends on which model you pick.\n\n| Limit | Cap |\n|-------|-----|\n| Per 5 hours | $12 |\n| Per week | $30 |\n| Per month | $60 |\n\nIf you have credits on your OpenCode Zen balance, you can enable the \"Use balance\" option in the console. When Go limits run out, it falls back to your Zen balance instead of blocking requests.\n\nCheaper models stretch further. Estimated request counts from the [official Go docs](https://opencode.ai/docs/go/):\n\n| Model | Requests per 5 hours | Requests per week | Requests per month |\n|-------|---------------------|-------------------|-------------------|\n| DeepSeek V4 Flash | 31,650 | 79,050 | 158,150 |\n| MiMo-V2.5 | 30,100 | 75,200 | 150,400 |\n| Hy3 | 4,300 | 10,750 | 21,500 |\n| Qwen3.7 Plus | 4,300 | 10,800 | 21,600 |\n| MiniMax M2.7 | 3,400 | 8,500 | 17,000 |\n| DeepSeek V4 Pro | 3,450 | 8,550 | 17,150 |\n| Qwen3.6 Plus | 3,300 | 8,200 | 16,300 |\n| MiMo-V2.5-Pro | 3,250 | 8,150 | 16,300 |\n| MiniMax M3 | 3,200 | 8,000 | 16,000 |\n| Kimi K2.7 Code | 1,350 | 3,380 | 6,750 |\n| Kimi K2.6 | 1,150 | 2,880 | 5,750 |\n| Qwen3.7 Max | 950 | 2,390 | 4,770 |\n| GLM-5.2 | 880 | 2,150 | 4,300 |\n| GLM-5.1 | 880 | 2,150 | 4,300 |\n| Grok 4.5 | 120 | 300 | 600 |\n| Kimi K3 | 110 | 250 | 490 |\n\nLean hard on Grok 4.5 or Kimi K3 and you will hit the 5-hour wall. Stick to DeepSeek V4 Flash, MiMo-V2.5, or MiniMax M2.7 and I doubt you hit anything unless agents run all day.\n\n## Setting it up\n\nThe setup takes about two minutes.\n\n1. Go to [opencode.ai/auth](https://go.bitdoze.com/opencode-go) and create an account\n2. Subscribe to Go and copy your API key\n3. In OpenCode, run `/connect`\n4. Select `OpenCode Go` and paste your key\n5. Run `/models` to see the available models\n\nThe base URL for the API is:\n\n```\nhttps://opencode.ai/zen/go/v1/chat/completions\n```\n\n### Using Go with other agents\n\nSince the endpoint is OpenAI-compatible, you can use Go with any tool that supports custom API endpoints:\n\n\n\n```bash\nexport OPENCODE_API_KEY=your-go-key\npi\n# /model, select opencode-go provider\n```\nSee the [Pi setup guide](/pi-coding-agent-setup-guide/) for full instructions.\n\n\n```bash\necho \"OPENAI_BASE_URL=https://opencode.ai/zen/go/v1/chat/completions\" >> ~/.hermes/.env\necho \"OPENAI_API_KEY=your-go-key\" >> ~/.hermes/.env\nhermes config set model opencode-go/glm-5.1\n```\nSee the [Hermes setup guide](/hermes-agent-setup-guide/) for full instructions.\n\n\n```typescript\nimport { createOpenAI } from '@ai-sdk/openai';\n\nconst opencodeGo = createOpenAI({\n baseURL: 'https://opencode.ai/zen/go/v1',\n apiKey: process.env.OPENCODE_API_KEY,\n});\n\nconst agent = new Agent({\n name: 'assistant',\n model: opencodeGo('glm-5.1'),\n});\n```\nSee the [Mastra guide](/build-ai-agent-mastra/) for a full walkthrough.\n\n\n```python\nfrom agno.agent import Agent\nfrom agno.models.openai import OpenAI\n\nagent = Agent(\n model=OpenAI(\n id=\"glm-5.1\",\n api_key=os.environ[\"OPENCODE_API_KEY\"],\n base_url=\"https://opencode.ai/zen/go/v1\",\n ),\n markdown=True,\n)\nagent.print_response(\"Your task here\")\n```\n\n\nSet the base URL to `https://opencode.ai/zen/go/v1/chat/completions` and use your Go API key in the OpenClaw config.\n\n\nAdd to `~/.codex/config.toml`:\n```toml\nmodel = \"glm-5.1\"\nmodel_provider = \"opencode-go\"\n\n[model_providers.opencode-go]\nname = \"OpenCode Go\"\nbase_url = \"https://opencode.ai/zen/go/v1\"\nenv_key = \"OPENCODE_API_KEY\"\nwire_api = \"chat\"\n```\n\n\n\n## Who this is for\n\nGo makes sense if:\n\n- You do not want to manage multiple API keys and billing dashboards\n- You are based outside the US and want low-latency access to models (the Singapore and EU endpoints help)\n- You want a predictable monthly cost instead of per-token billing\n- You use multiple agents (OpenCode + Pi + Hermes + Mastra + Agno) and want one key for all of them\n\nGo does not make sense if:\n\n- You only use one or two models and already have cheap API access\n- You need models not on the list (like Claude or GPT-4)\n- You are doing heavy production work that needs unlimited or very high limits\n\n## How it compares to managing your own keys\n\nI ran the numbers for my own usage. I typically use MiniMax M2.7 and DeepSeek V4 Pro through their direct APIs. My monthly cost with direct API keys was around $15-20 depending on how much I used agents that month.\n\nWith Go at $10/month, I get access to those same models plus others I would not have bothered setting up. The limits are generous enough that I have not hit them once in three weeks of daily use.\n\nThe trade-off is control. With direct API keys, I can switch providers if one goes down. With Go, I am relying on the OpenCode team to handle failover. In practice, I have not had any downtime issues, but it is worth noting.\n\n\nOpenRouter is a pay-per-token aggregator. You load credits and pay for what you use. Go is a flat monthly subscription with usage caps. If you use a lot of tokens, Go is cheaper. If you use very few tokens, OpenRouter might be cheaper. Go also includes models from providers that are not on OpenRouter.\n\n\n\nYes. The Go endpoint is OpenAI-compatible. Any tool that supports custom OpenAI-compatible endpoints can use it. This includes Hermes Agent, OpenClaw, Pi Agent, Mastra, and others. Set the base URL to `https://opencode.ai/zen/go/v1/chat/completions` and use your Go API key.\n\n\n\nRequests get blocked until the limit resets. The 5-hour limit resets on a rolling basis. If you have credits on your Zen balance, you can enable the \"Use balance\" option to keep going after hitting Go limits.\n\n\n\nOpenCode says providers on Go use zero retention and do not train on your prompts. That is their policy statement; I have not independently audited it.\n\n\n## My take after three weeks\n\nIf you already have cheap direct keys and only use one model, Go is optional. Nice, not life-changing. For me the win is simplicity: one invoice, one key, and it works in every agent I already run.\n\nIf you are still pasting five API keys around, start with the $5 first month. I leave Grok 4.5 and Kimi K3 for the hard stuff, MiniMax M3 or GLM-5.2 for normal agent sessions, and DeepSeek Flash or MiMo when I just need volume.\n\n\n\n## Related articles\n\n- [Build Your Own AI Agent with Mastra](/build-ai-agent-mastra/) — full guide using OpenCode Go with Mastra\n- [How to Use the Codex App with Any Model](/codex-app-any-model/) — run Go models inside the Codex app\n- [OpenCode Setup Guide: Install and Configure on a VPS](/opencode-setup-guide/) — full installation walkthrough\n- [OpenCode vs Pi Agent: Which Terminal Coding Agent to Use](/opencode-vs-pi-agent/) — side-by-side comparison\n- [Best Cheap Models for AI Coding Agents](/best-cheap-models-hermes-agent/) — model pricing and benchmarks\n- [GitHub Copilot Alternatives After the June 2026 Pricing Change](/github-copilot-alternatives-2026/) — what to switch to\n"}}}}