fix(chat): forward model options including reasoning to API requests

[subtleGradient]

Summary

Fixes an issue where model options passed as the second argument to the provider callable were not being forwarded to the Responses API request.

Problem

When users pass reasoning options:

const model = openrouter('google/gemini-3-flash-preview', {
  reasoning: {
    enabled: true,
    effort: 'medium',
  },
});

The reasoning config was stored in this.settings.modelOptions but never forwarded to the API request. The model would not return reasoning output.

Solution

In openrouter-chat-language-model.ts, extract model options from settings and forward them to the request params in both doGenerate() and doStream():

• reasoning (enabled, effort, maxTokens, summary)
• provider (routing config)
• models (fallback model IDs)
• transforms (message transforms)
• plugins (OpenRouter plugins)
• route (routing strategy)

Testing

Added 6 new tests covering model options forwarding scenarios:

• Reasoning options in doGenerate
• Reasoning options in doStream
• Provider routing options
• Fallback models
• Transforms
• No model options when not provided

All 134 tests pass ✅
Typecheck clean ✅

Related

• gap-916-reasoning-options-not-forwarded

[Copilot]

Pull request overview

This PR fixes an issue where model options (reasoning, provider, models, transforms, plugins, route) passed as the second argument to the provider callable were not being forwarded to OpenRouter API requests. The fix extracts these options from settings.modelOptions and includes them in the request parameters for both streaming and non-streaming calls.

Key changes:

• Forward model options including reasoning configuration to API requests
• Add duplicate forwarding logic in both doGenerate() and doStream() methods
• Include 6 new tests covering reasoning, provider routing, fallback models, and transforms

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
src/chat/openrouter-chat-language-model.ts	Adds model options extraction and forwarding logic (lines 85-86, 103-121, 309-310, 327-345) to include reasoning, provider, models, transforms, plugins, and route in API requests
src/tests/chat/openrouter-chat-language-model.test.ts	Adds mock SDK setup and 6 tests verifying model options are correctly included in request body for reasoning, provider, models, and transforms

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[subtleGradient]

should fix this: #307 (comment)
cc @cpakken @idriss

[ldriss]

@subtleGradient Thank you so much

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/OpenRouterTeam/ai-sdk-provider/@openrouter/ai-sdk-provider@321

commit: b025a83

[subtleGradient]

cc @cpakken @idriss

try #321 (comment) snapshot and lemme know if it solves your problem or no?

[cpakken]

cc @cpakken @idriss

try #321 (comment) snapshot and lemme know if it solves your problem or no?

No, it works with ai v6 generateText but when used with streamText it produces zod errors

https://github.com/cpakken/openrouter-v6-issue

here is a minimal reproduction.

My use case is using gemini-3-flash-preview and streaming object output and image input. But with the latest openrouter provider v1.5.4 using with ai v6, is fails to emit onFinish and fails to resolve the stream with reasoning turned on. Hoping this new openrouter provider with ai v6 will fix the issue