From ce3633d69e12670c24217244f88a3e4860aff94c Mon Sep 17 00:00:00 2001
From: iskandarsulaili <admin@cakobox.com>
Date: Fri, 21 Nov 2025 07:06:52 +0800
Subject: [PATCH] refactor: Conditionally include Gemini thinking level and
 refine Azure provider model info retrieval, client initialization, and usage
 chunk handling.

---
 src/api/providers/azure.ts     | 34 ++++++++++++++++++++++------------
 src/api/transform/reasoning.ts |  5 ++++-
 2 files changed, 26 insertions(+), 13 deletions(-)

diff --git a/src/api/providers/azure.ts b/src/api/providers/azure.ts
index a5dab910d8..d0b904330d 100644
--- a/src/api/providers/azure.ts
+++ b/src/api/providers/azure.ts
@@ -48,12 +48,10 @@ export class AzureHandler extends BaseProvider implements SingleCompletionHandle
 			const { default: AnthropicFoundry } = await import("@anthropic-ai/foundry-sdk")
 			const baseURL = this.options.azureBaseUrl || "https://your-endpoint.services.ai.azure.com/anthropic/"
 			const apiKey = this.options.azureApiKey || this.options.apiKey || "not-provided"
-			const apiVersion = this.options.azureApiVersion || "2023-06-01"
 
 			this.claudeClient = new AnthropicFoundry({
 				apiKey,
 				baseURL,
-				apiVersion,
 			})
 		} catch (error) {
 			throw new Error("Failed to initialize Azure Claude client: " + (error as Error).message)
@@ -285,12 +283,14 @@ export class AzureHandler extends BaseProvider implements SingleCompletionHandle
 			}
 
 			if (chunk.usage) {
+				// Azure OpenAI may include cache usage properties
+				const usage = chunk.usage as any
 				yield {
 					type: "usage",
 					inputTokens: chunk.usage.prompt_tokens || 0,
 					outputTokens: chunk.usage.completion_tokens || 0,
-					cacheWriteTokens: chunk.usage.cache_creation_input_tokens || undefined,
-					cacheReadTokens: chunk.usage.cache_read_input_tokens || undefined,
+					cacheWriteTokens: usage.cache_creation_input_tokens || undefined,
+					cacheReadTokens: usage.cache_read_input_tokens || undefined,
 				}
 			}
 		}
@@ -305,14 +305,24 @@ export class AzureHandler extends BaseProvider implements SingleCompletionHandle
 		const id = modelId && modelId in azureModels ? (modelId as AzureModelId) : azureDefaultModelId
 		const info: ModelInfo = azureModels[id]
 
-		const params = getModelParams({
-			format: this.isClaudeModel(id) ? "anthropic" : "openai",
-			modelId: id,
-			model: info,
-			settings: this.options,
-		})
-
-		return { id, info, ...params }
+		// Handle Claude and GPT models separately to maintain type safety
+		if (this.isClaudeModel(id)) {
+			const params = getModelParams({
+				format: "anthropic",
+				modelId: id,
+				model: info,
+				settings: this.options,
+			})
+			return { id, info, ...params }
+		} else {
+			const params = getModelParams({
+				format: "openai",
+				modelId: id,
+				model: info,
+				settings: this.options,
+			})
+			return { id, info, ...params }
+		}
 	}
 
 	async completePrompt(prompt: string): Promise<string> {
diff --git a/src/api/transform/reasoning.ts b/src/api/transform/reasoning.ts
index 07b3f80333..3f097f5af6 100644
--- a/src/api/transform/reasoning.ts
+++ b/src/api/transform/reasoning.ts
@@ -135,5 +135,8 @@ export const getGeminiReasoning = ({
 		return undefined
 	}
 
-	return { thinkingLevel: selectedEffort, includeThoughts: true }
+	return {
+		includeThoughts: true,
+		...(selectedEffort === "low" || selectedEffort === "high" ? { thinkingLevel: selectedEffort } : {}),
+	} as GeminiReasoningParams
 }