[PR #190] [CLOSED] fix: enable reasoning content preservation for OpenAI, GLM, and Qwen providers #225

New Issue

2026-06-06T22:09:46-04:00

yindo commented

2026-06-06 22:09:46 -04:00

📋 Pull Request Information

Original PR: https://github.com/vxcontrol/pentagi/pull/190
Author: @mason5052
Created: 3/10/2026
Status: ❌ Closed

Base: master ← Head: fix/glm-qwen-preserve-reasoning

📝 Commits (2)

a7180da fix: enable reasoning content preservation for GLM and Qwen providers
6fb3873 fix: also enable reasoning content preservation for OpenAI provider

📊 Changes

3 files changed (+3 additions, -0 deletions)

View changed files

📝 backend/pkg/providers/glm/glm.go (+1 -0)
📝 backend/pkg/providers/openai/openai.go (+1 -0)
📝 backend/pkg/providers/qwen/qwen.go (+1 -0)

📄 Description

Summary

OpenAI, GLM, and Qwen providers have models that support thinking/reasoning mode but their OpenAI client initialization was missing the WithPreserveReasoningContent() option. DeepSeek and Kimi providers already have this option enabled. This inconsistency could cause reasoning content to be silently dropped during inference when using thinking-capable models.

Changes

Added openai.WithPreserveReasoningContent() to openai.New() call in backend/pkg/providers/openai/openai.go
Added openai.WithPreserveReasoningContent() to openai.New() call in backend/pkg/providers/glm/glm.go
Added openai.WithPreserveReasoningContent() to openai.New() call in backend/pkg/providers/qwen/qwen.go

This makes all openai.New()-based providers consistent with DeepSeek and Kimi.

Provider Comparison (Before / After)

Provider	Thinking Models	WithPreserveReasoningContent (Before)	After
DeepSeek	deepseek-reasoner	Yes	Yes
Kimi	kimi-k2.5, kimi-k2-thinking	Yes	Yes
OpenAI	o3, o3-pro, o1, gpt-5.2, +15 more	No	Yes
GLM	glm-5, glm-4.7-flashx, +9 more	No	Yes
Qwen	qwq-32b, qwen3-235b-a22b, +20 more	No	Yes

Test plan

Verified all three packages compile cleanly: go build ./pkg/providers/openai/ ./pkg/providers/glm/ ./pkg/providers/qwen/
Existing provider tests remain unaffected
Manual verification: reasoning output should be preserved when using thinking-capable models

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/vxcontrol/pentagi/pull/190 **Author:** [@mason5052](https://github.com/mason5052) **Created:** 3/10/2026 **Status:** ❌ Closed **Base:** `master` ← **Head:** `fix/glm-qwen-preserve-reasoning` --- ### 📝 Commits (2) - [`a7180da`](https://github.com/vxcontrol/pentagi/commit/a7180da2811a36c5b2b080d9369da4c338719fb4) fix: enable reasoning content preservation for GLM and Qwen providers - [`6fb3873`](https://github.com/vxcontrol/pentagi/commit/6fb3873943ab2690b6560969938aa0ece92b58c6) fix: also enable reasoning content preservation for OpenAI provider ### 📊 Changes **3 files changed** (+3 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `backend/pkg/providers/glm/glm.go` (+1 -0) 📝 `backend/pkg/providers/openai/openai.go` (+1 -0) 📝 `backend/pkg/providers/qwen/qwen.go` (+1 -0) </details> ### 📄 Description ## Summary OpenAI, GLM, and Qwen providers have models that support thinking/reasoning mode but their OpenAI client initialization was missing the `WithPreserveReasoningContent()` option. DeepSeek and Kimi providers already have this option enabled. This inconsistency could cause reasoning content to be silently dropped during inference when using thinking-capable models. ## Changes - Added `openai.WithPreserveReasoningContent()` to `openai.New()` call in `backend/pkg/providers/openai/openai.go` - Added `openai.WithPreserveReasoningContent()` to `openai.New()` call in `backend/pkg/providers/glm/glm.go` - Added `openai.WithPreserveReasoningContent()` to `openai.New()` call in `backend/pkg/providers/qwen/qwen.go` This makes all `openai.New()`-based providers consistent with DeepSeek and Kimi. ## Provider Comparison (Before / After) | Provider | Thinking Models | WithPreserveReasoningContent (Before) | After | |----------|----------------|--------------------------------------|-------| | DeepSeek | deepseek-reasoner | Yes | Yes | | Kimi | kimi-k2.5, kimi-k2-thinking | Yes | Yes | | OpenAI | o3, o3-pro, o1, gpt-5.2, +15 more | **No** | Yes | | GLM | glm-5, glm-4.7-flashx, +9 more | **No** | Yes | | Qwen | qwq-32b, qwen3-235b-a22b, +20 more | **No** | Yes | ## Test plan - [ ] Verified all three packages compile cleanly: `go build ./pkg/providers/openai/ ./pkg/providers/glm/ ./pkg/providers/qwen/` - [ ] Existing provider tests remain unaffected - [ ] Manual verification: reasoning output should be preserved when using thinking-capable models --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

yindo added the pull-request label 2026-06-06 22:09:46 -04:00

yindo closed this issue

2026-06-06 22:09:47 -04:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: vxcontrol/pentagi#225