[BUG] kimi k2 thinking is not working on OpenAi compatible

[bozoweed]

Problem (one or two sentences)

all in title

Context (who is affected and when)

kimi k2 thinking model on openai provider

Reproduction steps

use open ai compatible provider
use kimi k2 thinking model
ask for some things

Expected result

work

Actual result

not wrok

Variations tried (optional)

No response

App Version

3.39.2

API Provider (optional)

OpenAI Compatible

Model Used (optional)

kimi k2 thinking

Roo Code Task Links (optional)

Relevant logs or errors (optional)

[roomote]

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

Root Cause Analysis

When using the "OpenAI Compatible" provider with Kimi K2 thinking model, the thinking output is not working because:

1. The OpenAiHandler in src/api/providers/openai.ts can parse reasoning_content from API responses (line 205-209), but does NOT send the thinking: { type: "enabled" } parameter in the request
2. Many thinking models (Kimi K2, DeepSeek Reasoner, GLM-4.7, etc.) require the thinking parameter to be explicitly enabled in the API request to return thinking/reasoning content
3. Currently there's no way for users to enable this parameter when using the OpenAI Compatible provider

Proposed Solution

Add a new setting openAiThinkingModeEnabled (similar to the existing openAiR1FormatEnabled option) that:

1. Adds the thinking parameter to requests - When enabled, includes thinking: { type: "enabled" } in API calls
2. Enables reasoning preservation - Sets preserveReasoning: true in model info so thinking content is maintained across multi-turn conversations
3. UI toggle - Add a checkbox in the OpenAI Compatible settings section

Files to modify:

1. packages/types/src/global-settings.ts - Add openAiThinkingModeEnabled setting
2. packages/types/src/provider-settings.ts - Add setting to schema
3. src/api/providers/openai.ts - Use the setting to add thinking parameter and handle response parsing
4. webview-ui/src/components/settings/providers/OpenAi.tsx - Add UI toggle
5. Add tests in src/api/providers/__tests__/openai.spec.ts

This approach:

• Is consistent with the existing openAiR1FormatEnabled pattern
• Gives users explicit control over thinking mode
• Works with various thinking models (Kimi K2, DeepSeek, GLM-4.7, etc.) via OpenAI Compatible provider

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

[ArchimedesCrypto]

@roomote Looks good to me. Approved. Consider edge cases in your tests and ensure they are all passing.

[roomote]

Thanks for the approval! I'm starting the implementation now. I'll open a PR to share progress shortly.

[roomote]

Opened a PR with my attempt at addressing this: #10613

The implementation adds a new "Enable thinking mode" setting for the OpenAI Compatible provider that sends the thinking: { type: "enabled" } parameter in API requests. This should work with thinking models like Kimi K2, DeepSeek Reasoner, and GLM-4.

Feedback welcome!