How to connect OpenCode to CheapTokens (opencode.json)

To connect OpenCode to CheapTokens, add a custom OpenAI-compatible provider: store your API key with /connect (provider ID cheaptokens), then describe the provider in opencode.json with the base URL https://api.cheaptokens.dev/v1. After that your CheapTokens models appear in /models.

This guide follows the OpenCode providers documentation (page updated October 20, 2026; checked October 22, 2026). The same approach works for the other tools in this series: Cline, Aider and Kilo Code.

What you need#

• OpenCode installed (see its documentation for your platform).
• A CheapTokens API key. Create one in your account and top up the balance; one balance covers all models. The basics are in the CheapTokens docs.
• The API facts:
  • base URL: https://api.cheaptokens.dev/v1
  • compatibility: OpenAI Chat Completions and Anthropic Messages
  • model id used below: deepseek-ai/DeepSeek-V4-Flash-0731 (380K context, up to 32,768 output tokens). Model ids are listed in the CheapTokens docs: names differ between providers, so copy the id from there.

Step 1. Store the key with /connect#

1. Start OpenCode in your project and run /connect.
2. Scroll to Other.
3. Enter a unique provider ID: cheaptokens.
4. Paste your CheapTokens API key.

This step only stores the credential. OpenCode still does not know the address or the models of the provider: that is what opencode.json is for.

Step 2. Describe the provider in opencode.json#

Create or edit opencode.json and add the provider. The provider ID in provider must match the one you entered in /connect.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "cheaptokens": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "CheapTokens",
      "options": {
        "baseURL": "https://api.cheaptokens.dev/v1"
      },
      "models": {
        "deepseek-ai/DeepSeek-V4-Flash-0731": {
          "name": "DeepSeek-V4-Flash",
          "limit": {
            "context": 380000,
            "output": 32768
          }
        }
      }
    }
  }
}

What each part does:

• npm — @ai-sdk/openai-compatible is the package for endpoints that use /v1/chat/completions, which is what CheapTokens exposes. (OpenCode uses @ai-sdk/openai for endpoints built on /v1/responses.)
• name — the label shown in the OpenCode interface.
• options.baseURL — the CheapTokens API address.
• models — keyed by model id; the name inside is a display name.
• limit.context / limit.output — optional; they tell OpenCode the model's context window and maximum answer length.

Keep the key in an environment variable instead (optional)#

If you prefer not to keep the key in OpenCode's credential store, add apiKey to options with the {env:VARIABLE} syntax the docs recommend:

"options": {
  "baseURL": "https://api.cheaptokens.dev/v1",
  "apiKey": "{env:CHEAPTOKENS_API_KEY}"
}

Then export CHEAPTOKENS_API_KEY in your shell. Do not commit a real key to the repository. If you are new to keys, read what an API key is and how to store it safely.

Step 3. Check that it works#

Run /models and pick DeepSeek-V4-Flash under the CheapTokens provider. Then send a short prompt, for example: "Summarize what this repository does in three sentences." A normal answer means the base URL, key and model id are all correct.

Which CheapTokens model to pick#

Add as many entries to models as you like — each is a block like the one above, with its own id.

For an OpenCode session, which sends many requests in a loop, DeepSeek-V4-Flash is a good default: output is cheaper than OpenRouter's reference price, answers can reach 32,768 tokens, and the price has no peak hours. MiniMax-M2.7 is also cheaper than OpenRouter's reference prices on both input and output. GLM-5.3-Flash's output is cheaper than OpenRouter's reference price as well. Current prices (checked live, no numbers to go stale):

To estimate a working session, for example 20 requests with 100,000 input and 5,000 output tokens each:

See all prices and the DeepSeek page for more.

If it doesn't work#

• The provider or models are missing from /models. Check that opencode.json is valid JSON (no trailing commas) and that the key under provider is exactly the ID you entered in /connect. If both are correct, restart OpenCode.
• 401 / authentication error. Re-run /connect and paste the key again, or check the CHEAPTOKENS_API_KEY variable in the shell where OpenCode starts.
• 404 or "model not found". The model id in models must match the CheapTokens id character for character, including the vendor prefix (deepseek-ai/…).
• Connection errors. The base URL must end with /v1 and contain no extra path. See common connection errors.
• "Not enough tokens". Top up the balance; see what the error means.

Next steps#

You now have OpenCode running on CheapTokens with one balance for all models. Get a key and make your first call from the CheapTokens docs.