[BUG] [OpenAI completion error] Invalid combination of reasoning_effort and thinking type: medium + disabled

[nicky-zs]

Problem (one or two sentences)

For the OpenAI Compatible Model, when reasoning effort is enabled and, for example, the "medium" model reasoning effort is selected, the LLM calling will fail:
OpenAI completion error: 400 Invalid combination of reasoning_effort and thinking type: medium + disabled

Context (who is affected and when)

For the OpenAI Compatible Model, when reasoning effort is enabled and, for example, the "medium" model reasoning effort is selected, the LLM calling will fail.

Reproduction steps

1. OpenAI Compatible Model
2. Enable Reasoning Effort
3. "medium" or else
4. volcengine: https://ark.cn-beijing.volces.com/api/coding/v3: ark-code-latest used.

Expected result

success LLM call

Actual result

OpenAI completion error: 400 Invalid combination of reasoning_effort and thinking type: medium + disabled

Variations tried (optional)

No response

App Version

3.43.0 (d8949f9)

API Provider (optional)

None

Model Used (optional)

No response

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote]

I've analyzed this issue and here's my proposed implementation plan:

Root Cause

When using the OpenAI Compatible provider with reasoning effort enabled (e.g., "medium"), the error occurs because some OpenAI-compatible APIs (like volcengine's Ark API) require both:

1. The reasoning_effort parameter (e.g., "medium")
2. A thinking parameter set to { type: "enabled" }

Currently, src/api/providers/openai.ts (the OpenAI Compatible handler) sends reasoning_effort but does NOT add the thinking parameter. This creates an invalid state where reasoning_effort is "medium" but thinking type remains "disabled" on the API side.

The base-openai-compatible-provider.ts already handles this correctly by checking model.supportsReasoningBinary and adding the thinking parameter accordingly, but openai.ts lacks this logic.

Proposed Fix

Modify src/api/providers/openai.ts to add thinking: { type: "enabled" } when:

1. Reasoning effort is being used (the reasoning object from getModelParams() is non-empty), AND
2. The model's supportsReasoningBinary flag is true (set via custom model info)

Changes:

1. In src/api/providers/openai.ts (createMessage method, around line 161):

   • Add logic to include thinking: { type: "enabled" } in the request when reasoning is present AND modelInfo.supportsReasoningBinary is true

2. Also update the O3 family message handler if needed (around line 346), following the same pattern

This approach:

• Is consistent with how base-openai-compatible-provider.ts handles this scenario
• Uses the existing supportsReasoningBinary model capability flag
• Users can enable this by setting supportsReasoningBinary: true in their custom model info
• Won't affect existing configurations that don't need the thinking parameter

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.

[nicky-zs]

@roomote approve

[roomote]

Thanks for the approval! I'll proceed with implementing the fix now. I'll open a PR to share progress once the changes are ready.

[roomote]

Opened a draft PR with my attempt at addressing this issue: #11032

The fix adds the thinking: { type: "enabled" } parameter when both:

1. Reasoning effort is being used
2. The model's supportsReasoningBinary flag is true

To use this fix, set supportsReasoningBinary: true in your custom model info configuration.

Feedback welcome!