StepFun Step 3.7 Flash + OpenCode: Cheapest Reasoning Setup

Step 3.7 Flash is an open 198B mixture-of-experts vision-language model with about 11B active parameters per token. It accepts text and images, has a 256K context window, and offers low, medium, and high reasoning levels. OpenCode can use it through an OpenAI-compatible provider.

This is the OpenCode + StepFun setup. For OpenCode’s general provider model, see OpenCode custom providers.

Step 1: Install OpenCode and get a StepFun key#

npm install -g opencode-ai

Create a StepFun API key. Native Windows and WSL both work with OpenCode. See install OpenCode on Windows with WSL when the project uses Linux tooling.

Step 2: Add the StepFun provider#

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "stepfun": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "StepFun",
      "options": {
        "baseURL": "https://api.stepfun.com/v1",
        "apiKey": "{env:STEPFUN_API_KEY}"
      },
      "models": {
        "step-3.7-flash": { "name": "Step 3.7 Flash" },
        "step-3.5-flash": { "name": "Step 3.5 Flash" }
      }
    }
  },
  "model": "stepfun/step-3.7-flash"
}

Export the key:

export STEPFUN_API_KEY="your-stepfun-key"

Confirm the exact base URL and model IDs on StepFun’s docs.

Step 3: Run it#

opencode

The default is stepfun/step-3.7-flash. For the lowest cost on simpler tasks, switch to step-3.5-flash.

Pricing#

Requesty’s live catalog lists Step 3.7 Flash at $0.20 input and $1.15 output per million tokens before its displayed discount. Check StepFun’s account console for the direct rate attached to your plan.

Troubleshooting#

• If the provider is missing, make sure the stepfun key matches the model prefix.
• For authentication errors, confirm that STEPFUN_API_KEY is set in the launching shell.
• If the model is rejected, verify its ID in StepFun’s live model list.
• For format errors, confirm that the OpenAI-compatible endpoint is selected.

Verify the provider#

Restart OpenCode and ask it to read one file without editing it. Confirm the active provider and model before approving commands.

For the broader budget picture, see cheapest AI coding API in 2026.