Cost tracking shows $0.00 for custom providers using @ai-sdk/openai-compatible even when cost fields are configured

[fabienbarbaud]

Description

When using a custom provider configured via opencode.json with @ai-sdk/openai-compatible, the UI does not track or display $ Spent. The cost remains at $0.00 regardless of token usage, even when cost fields are explicitly configured per the config schema.
This is a similar issue to what was reported on the popular fork anomalyco/opencode#17223.

Plugins

No response

OpenCode version

1.14.22

Steps to reproduce

1. Configure a custom provider in opencode.json with @ai-sdk/openai-compatible and cost fields specified:

{
  "provider": {
    "my-proxy": {
      "name": "My LiteLLM Proxy",
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://proxy.example.com/v1",
        "apiKey": "sk-..."
      },
      "models": {
        "my-model": {
          "name": "My Model",
          "cost": {
            "input": 5.5e-07,
            "output": 2.75e-06,
            "cache": {
              "read": 5.5e-08,
              "write": 6.875e-07
            }
          },
          "limit": {
            "context": 200000,
            "output": 64000
          }
        }
      }
    }
  }
}

2. Start a session using the custom provider model.
3. Send messages and observe the cost display in the status bar.

Expected Behavior

The UI should calculate and display a running cost using either:

• The cost values defined in the model config combined with the token usage from the API response (usage.prompt_tokens, usage.completion_tokens), OR
• The cost header returned by the proxy (e.g., x-litellm-response-cost).

Actual Behavior

• $ Spent remains at $0.00 for the entire session.
• The cost config fields on custom provider models have no observable effect.
• The upstream API (LiteLLM proxy) correctly returns token usage in the response body:

usage: {
    prompt_tokens: 10,
    completion_tokens: 52,
    total_tokens: 62
  }

• The upstream API also returns a cost header: x-litellm-response-cost: 0.0002015
• OpenCode appears to neither calculate cost from token counts nor read the x-litellm-response-cost header.

Screenshot and/or share link

No response

Operating System

Ubuntu 24

Terminal

No response

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• Cost tracking ($ Spent) does not work for custom provider models #17223: Cost tracking ($ Spent) does not work for custom provider models — same symptom of $0.00 cost display for custom providers (you also reference this issue in your report)

If this is indeed the same underlying issue, please consider adding your specific details (the @ai-sdk/openai-compatible configuration and x-litellm-response-cost header context) as a comment on #17223 instead.

[maurerle]

Yes, duplicate of #17223:
Your values in the config are too low to create meaningful steps.
Try 0.3 as a cost for example, which is in $ per 1M tokens.

[fabienbarbaud]

Update — confirmed fixed on my end.

@maurerle was right about the cost scale — my values were per-token instead of per-1M-tokens. That fixed the $0.00 display.
However, I also had two related config gaps that left the UI broken even after fixing costs:

1. limit.context is required for the % usage display

OpenCode uses model.limit.context for the percentage calculation, not model.context_window. Without it, context always shows 0% regardless of actual token usage.

2. includeUsage: true is needed in provider options

When using @ai-sdk/openai-compatible with a LiteLLM proxy, streaming responses don't include token counts by default. Both cost and context tracking fail silently without this flag.

Full working config excerpt

{
  "provider": {
    "litellm": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://proxy.example.com/v1",
        "includeUsage": true
      },
      "models": {
        "my-model": {
          "limit": { "context": 202000, "output": 25344 },
          "cost": { "input": 1.4, "output": 4.4 }
        }
      }
    }
  }
}

All three (cost scale, limit.context, includeUsage) were required to get cost tracking and context percentage working correctly.

---

Closing this as resolved — the underlying behavior could use better docs, but it's working as designed once configured properly.