fix(provider): guard reasoningSummary injection for @ai-sdk/openai-compatible providers

[nazarhnatyshen]

Issue for this PR

Closes #21237
Closes #18882

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

The bug: When using @ai-sdk/openai-compatible provider (e.g. LiteLLM proxy) with any GPT-5.x model, opencode injects reasoningSummary: "auto" into every request. LiteLLM does not recognize this parameter and rejects the request:

litellm.BadRequestError: OpenAIException - Unknown parameter: 'reasoningSummary'

Why it happens: In options() (transform.ts:835), reasoningSummary is set unconditionally for any provider whose model.api.id contains "gpt-5". The variants() function in the same file already correctly guards reasoningSummary behind a provider check — only setting it for @ai-sdk/openai, @ai-sdk/azure, and @ai-sdk/github-copilot. The options() function lacks this guard.

The fix: Apply the same provider check in options() as already exists in variants(). reasoningSummary is now only injected for the three providers that translate it into the OpenAI Responses API format (reasoning.summary). For @ai-sdk/openai-compatible, the parameter is skipped entirely, which is correct — compatible proxies pass options directly in the request body without any translation.

How did you verify your code works?

Reproduced the error locally using a LiteLLM proxy with gpt-5.4 configured as an @ai-sdk/openai-compatible model. After this fix, the request succeeds without the Unknown parameter: 'reasoningSummary' error. Models on native @ai-sdk/openai continue to send reasoningSummary: "auto" as before.

Screenshots / recordings

N/A — no UI change

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Potential Duplicate PRs Found

1. PR fix: OpenCode was injecting reasoningSummary for GPT-5 and forwarding it through the openai-compatible chat adapter #14783: "fix: OpenCode was injecting reasoningSummary for GPT-5 and forwarding it through the openai-compatible chat adapter"

   • fix: OpenCode was injecting reasoningSummary for GPT-5 and forwarding it through the openai-compatible chat adapter #14783
   • Why related: Directly addresses the same issue - injecting reasoningSummary for GPT-5 through openai-compatible adapter, causing forwarding errors

2. PR fix(opencode): guard reasoningSummary for azure in options() #21575: "fix(opencode): guard reasoningSummary for azure in options()"

   • fix(opencode): guard reasoningSummary for azure in options() #21575
   • Why related: Similar fix that guards reasoningSummary injection for specific providers in the options() function

3. PR fix(provider): avoid incompatible GPT-5 defaults and sanitize Gemini tool schemas #18915: "fix(provider): avoid incompatible GPT-5 defaults and sanitize Gemini tool schemas"

   • fix(provider): avoid incompatible GPT-5 defaults and sanitize Gemini tool schemas #18915
   • Why related: Addresses incompatible GPT-5 defaults across different providers, similar scope

These PRs address the same underlying problem of guarding provider-specific parameters like reasoningSummary in the options() function for GPT-5 models.

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍

[rekram1-node]

clean fix thxx