[FEATURE]: Use models.dev data for custom providers

[mcowger]

Feature hasn't been suggested before.

• [x] I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Currently, when using preconfigured providers (like Anthropic, OpenAI, etc.) with a whitelist, OpenCode automatically fetches model metadata (pricing, context limits, capabilities) from models.dev:

{
  "provider": {
    "anthropic": {
      "whitelist": [
        "claude-sonnet-4-5",
        "claude-opus-4-5"
      ]
    }
  }
}

However, for custom providers, all model metadata must be manually specified:

{
  "provider": {
    "customprovider": {
      "name": "AI Home",
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "apiKey": "{env:AIHOME_API_KEY}",
        "baseURL": "https://api.example.com"
      },
      "models": {
        "minimax/minimax-m2.1": {
          "limit": { "context": 1000000, "output": 32000 },
          "cost": { "input": 0.3, "output": 1.2 },
          "temperature": true,
          "reasoning": true,
          "tool_call": true
        }
      }
    }
  }
}

This is tedious, error-prone, and requires tracking pricing/limit changes manually.

Proposed Solution

Add a use_models_dev option that allows custom providers to reference model metadata from models.dev:

{
  "provider": {
    "customprovider": {
      "name": "AI Home",
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "apiKey": "{env:API_KEY}",
        "baseURL": "https://api.example.com"
      },
      "models": {
        "minimax/minimax-m2.1": {
          "use_models_dev": "openrouter/minimax/minimax-m2.1"
        }
      }
    }
  }
}

This would automatically import:

• Context/input/output limits
• Pricing (input, output, cache read/write)
• Capabilities (tool calls, reasoning, temperature, attachments)
• Modalities (text, image, audio, video, PDF support)
• Variants and other metadata

Benefits

1. Less Configuration - No need to manually track pricing and limits
2. Auto-Updates - When models.dev updates pricing or limits, OpenCode automatically uses latest values
3. Accuracy - Ensures metadata matches the source provider
4. Flexibility - Can still override specific fields when needed

User Override Support

Users should be able to override imported metadata:

{
  "models": {
    "minimax/minimax-m2.1": {
      "use_models_dev": "openrouter/minimax/minimax-m2.1",
      "limit": {
        "context": 500000  // Override imported value
      }
    }
  }
}

Merge priority: user config > models.dev import > defaults

Error Handling

If the specified use_models_dev model ID is not found in models.dev:

• Log a warning with the invalid ID
• List available providers to help user find correct ID
• Fall back to any manually specified metadata in config
• Don't crash or fail config loading

Finding Model IDs

Model IDs follow the format: provider/model-name Examples:

• openrouter/minimax/minimax-m2.1
• openrouter/anthropic/claude-3.7-sonnet
• anthropic/claude-opus-4-5 Users can find valid IDs at https://models.dev/

Use Cases

• Proxy providers - Using OpenRouter, AI Gateway, or other proxies that support models already in models.dev
• Self-hosted deployments - Running models locally that match commercial offerings
• Testing - Quickly switching between providers without reconfiguring metadata
• Multi-provider setups - Accessing the same model through different endpoints

Example: Real-world scenario

A user wants to access Minimax M2.1 through their self-hosted AI proxy. Instead of manually looking up and copying all the metadata, they can simply reference the OpenRouter version:

{
  "provider": {
    "my-ai-proxy": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "apiKey": "{env:MY_PROXY_KEY}",
        "baseURL": "https://my-proxy.internal/v1"
      },
      "models": {
        "minimax-m2.1": {
          "use_models_dev": "openrouter/minimax/minimax-m2.1"
        }
      }
    }
  }
}

All pricing, limits, and capabilities are automatically imported from models.dev.

Non-Goals

• This feature does NOT add new models to models.dev
• This feature does NOT validate that the provider actually supports the model
• This feature does NOT change how preconfigured providers work.