[BUG] options.reasoningEffort not passed to API calls for custom @ai-sdk/openai provider

[hhieuu]

Description

Starting from v1.3.4, options.reasoningEffort plus options.reasoningSummary, and by extension variants.<name>.reasoningEffort plus variants.<name>.reasoningSummary, configured in opencode.json is no longer passed on to API calls when @ai-sdk/openai provider is used as the NPM SDK for a custom provider. The option still works with v1.3.3, so I believe this regression is introduced with the move to AI SDK v6.

The documentation for the OpenAI SDK supports the use of reasoningEffort and reasoningSummary: docs here. This effectively make GPT 5.4 and GPT 5.3 Codex on custom provider mostly unusable.

Plugins

No response

OpenCode version

1.3.13, but this regression appears first on 1.3.4 and forward.

Steps to reproduce

1. Create a custom provider in config against a custom Responses API compatible provider. Example:

{
  "provider": {
    "my_provider": {
      "my_model": {
        "npm": "@ai-sdk/openai provider",
        "name": "My Custom Provider",
        "options": {
          "baseURL": "https://custom-openai-compatible-url/v1",
          "apiKey": "sk-my-api-key"
        },
        "models": {
          "gpt-5.3-codex": {
            "name": "gpt-5.3-codex",
            "options": {
              "include": ["reasoning.encrypted_content"], // included in request body
              "reasoningEffort": "high", // NOT included in request body
              "reasoningSummary": "auto", // NOT included in request body
              "textVerbosity": "medium" // included in request body
            }
          },
          "gpt-5.4": {
            "name": "gpt-5.3-codex",
            "options": {
              "include": ["reasoning.encrypted_content"] // included in request body
            },
            "variants": {
              "high": {
                "reasoningEffort": "high", // NOT included in request body
                "reasoningSummary": "auto", // NOT included in request body
                "textVerbosity": "medium" // included in request body
              }
            }
          }
        }
      }
    }
  }
}

2. Make a simple request in OpenCode with either model.
3. Intercept and inspect the request body

Expect: Reasoning effort and summary is included in the request body as below.

{
  ...
  "reasoning": {
    "effort": "high",
    "summary": "auto"
  },
  ...
}

Observed: "reasoning" is not added into the request body -> bug.

Screenshot and/or share link

reasoningEffort and reasoningSummary usage docs here

Operating System

Linux

Terminal

No response

[github-actions]

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

• Azure provider: reasoningEffort model option silently ignored #20275: Similar issue where reasoningEffort model option is silently ignored for a custom provider (Azure variant), with root cause analysis pointing to transform.ts and provider option key mismatches — the underlying mechanism may be the same for @ai-sdk/openai.

[hhieuu]

To add, this actually does not happen with the @ai-sdk/openai-compatible provider that uses the Chat Completion API. With Chat Comp API, reasoningEffort in config is correctly passed on as reasoning_effort.