fix: respect 1M token context window for auto-condensing

- Update Task.ts to use actual context window from api.getModel().info
- This ensures the 1M token extension is properly respected when enabled
- Fixes auto-condensing triggering at base model limits instead of extended limits

Fixes #9831

Author: roomote

src/core/task/Task.ts

@@ -3338,10 +3338,12 @@ export class Task extends EventEmitter<TaskEvents> implements TaskLike {
 		const { profileThresholds = {} } = state ?? {}
 
 		const { contextTokens } = this.getTokenUsage()
-		const modelInfo = this.api.getModel().info
+		// Get the actual model info from the API handler which includes 1M context updates
+		const actualModel = this.api.getModel()
+		const modelInfo = actualModel.info
 
 		const maxTokens = getModelMaxOutputTokens({
-			modelId: this.api.getModel().id,
+			modelId: actualModel.id,
 			model: modelInfo,
 			settings: this.apiConfiguration,
 		})
@@ -3483,10 +3485,12 @@ export class Task extends EventEmitter<TaskEvents> implements TaskLike {
 		const { contextTokens } = this.getTokenUsage()
 
 		if (contextTokens) {
-			const modelInfo = this.api.getModel().info
+			// Get the actual model info from the API handler which includes 1M context updates
+			const actualModel = this.api.getModel()
+			const modelInfo = actualModel.info
 
 			const maxTokens = getModelMaxOutputTokens({
-				modelId: this.api.getModel().id,
+				modelId: actualModel.id,
 				model: modelInfo,
 				settings: this.apiConfiguration,
 			})