feat: Add GPT-5 Pro with background mode auto-resume and polling

[hannesrudolph]

Summary

Adds GPT-5 Pro model with OpenAI Responses API background mode support. Background requests can take several minutes, so this implements resilient streaming with automatic recovery.

Why

GPT-5 Pro is a slow, reasoning-focused model that can take several minutes to respond. The standard streaming approach times out or appears stuck. OpenAI's Responses API background mode is designed for these long-running requests.

What Changed

Model Addition

• Added gpt-5-pro-2025-10-06 with backgroundMode: true flag in model metadata

Background Mode Implementation

• Auto-enables background: true, stream: true, store: true for flagged models
• Emits status events: queued → in_progress → completed/failed
• Shows status labels in UI spinner ("background mode (queued)…", etc.)

Resilient Streaming

• Auto-resume: If stream drops, resumes from last sequence number using GET /v1/responses/{id}?starting_after={seq}
• Poll fallback: If resume fails after 3 retries, polls every 2s until completion (up to 20 minutes)
• Synthesizes final output and usage data when polling completes

Files Changed

• packages/types/src/providers/openai.ts - Model metadata
• src/api/providers/openai-native.ts - Background mode logic, auto-resume, polling
• src/core/task/Task.ts - Status event handling
• webview-ui/src/utils/backgroundStatus.ts - Status label mapping
• webview-ui/src/components/chat/* - UI status display

Testing

• Background mode status emission and lifecycle
• Auto-resume on stream drop with exponential backoff
• Poll fallback when resume exhausts retries
• Usage tracking parity with non-background requests
• UI status label mapping

---

Important

Adds GPT-5 Pro model with background mode, implementing resilient streaming and UI updates for status tracking.

• Behavior:
  • Adds gpt-5-pro-2025-10-06 model with backgroundMode: true in openai.ts.
  • Implements background mode with auto-resume and polling in openai-native.ts.
  • UI updates in ChatRow.tsx and ChatView.tsx for background status display.
• Resilient Streaming:
  • Auto-resume from last sequence number on stream drop.
  • Polls every 2s for up to 20 minutes if resume fails.
• UI:
  • Displays background mode status in chat UI with labels like "queued", "in_progress", etc.
  • Updates ChatRow.tsx and ChatView.tsx to handle new status labels.
• Testing:
  • Adds tests in backgroundStatus.spec.ts for status label mapping.
  • Tests for auto-resume and polling in openai-native.spec.ts.

^{This description was created by for 706138e. You can customize this summary. It will automatically update as commits are pushed.}

[hannesrudolph]

Any update? When will the pr merged?

Not sure yet. Will look at it after we get Native Tool Calling out later this week. Sorry for the delay.

[Lagyu]

Not sure yet. Will look at it after we get Native Tool Calling out later this week. Sorry for the delay.

@hannesrudolph Thank you so much for your effort. I am looking forward to use this feature soon in production :)

[hannesrudolph]

Not sure yet. Will look at it after we get Native Tool Calling out later this week. Sorry for the delay.

@hannesrudolph Thank you so much for your effort. I am looking forward to use this feature soon in production :)

Feel free to give it a try in its current form. I have not tested it since I just updated it and have to run out! BBL

[Lagyu]

@hannesrudolph Hi, I’ve been testing the gpt-5-pro branch with Azure OpenAI (Responses API, gpt-5-pro-2025-10-06 model and the same deployment name) and ran into an issue with background mode when openAiNativeBaseUrl already includes /v1 (e.g. https://<resource>.openai.azure.com/openai/v1/).

For long-running background responses, the initial responses.create call works, but when the stream is terminated and Roo tries to resume or poll, the URL becomes:

https://<resource>.openai.azure.com/openai/v1//v1/responses/{responseId}...

That extra /v1 causes Azure to return 404, so resume and polling both fail.

I tested the following change on src/api/providers/openai-native.ts locally and it fixes the issue for me with Azure while preserving the current behavior for https://api.openai.com:

@@
-	private normalizeUsage(usage: any, model: OpenAiNativeModel): ApiStreamUsageChunk | undefined {
+	private buildResponsesUrl(path: string): string {
+		const rawBase = this.options.openAiNativeBaseUrl || "https://api.openai.com"
+		// Normalize base by trimming trailing slashes
+		const normalizedBase = rawBase.replace(/\/+$/, "")
+		// If the base already ends with a version segment (e.g. /v1), do not append another
+		const hasVersion = /\/v\d+(?:\.\d+)?$/.test(normalizedBase)
+		const baseWithVersion = hasVersion ? normalizedBase : `${normalizedBase}/v1`
+		const normalizedPath = path.startsWith("/") ? path : `/${path}`
+		return `${baseWithVersion}${normalizedPath}`
+	}
+
+	private normalizeUsage(usage: any, model: OpenAiNativeModel): ApiStreamUsageChunk | undefined {
@@
-		const apiKey = this.options.openAiNativeApiKey ?? "not-provided"
-		const baseUrl = this.options.openAiNativeBaseUrl || "https://api.openai.com"
-		const url = `${baseUrl}/v1/responses`
+		const apiKey = this.options.openAiNativeApiKey ?? "not-provided"
+		const url = this.buildResponsesUrl("responses")
@@
-		const apiKey = this.options.openAiNativeApiKey ?? "not-provided"
-		const baseUrl = this.options.openAiNativeBaseUrl || "https://api.openai.com"
+		const apiKey = this.options.openAiNativeApiKey ?? "not-provided"
 		const resumeMaxRetries = this.options.openAiNativeBackgroundResumeMaxRetries ?? 3
 		const resumeBaseDelayMs = this.options.openAiNativeBackgroundResumeBaseDelayMs ?? 1000
@@
-				const resumeUrl = `${baseUrl}/v1/responses/${responseId}?stream=true&starting_after=${lastSeq}`
+				const resumeUrl = this.buildResponsesUrl(
+					`responses/${responseId}?stream=true&starting_after=${lastSeq}`,
+				)
@@
-				const pollRes = await fetch(`${baseUrl}/v1/responses/${responseId}`, {
+				const pollRes = await fetch(this.buildResponsesUrl(`responses/${responseId}`), {
 					method: "GET",
 					headers: {
 						Authorization: `Bearer ${apiKey}`,
 					},

Behavior-wise:

• openAiNativeBaseUrl = "https://api.openai.com" → URLs become https://api.openai.com/v1/responses/... (same as today).
• openAiNativeBaseUrl = "https://<resource>.openai.azure.com/openai/v1" → URLs become https://<resource>.openai.azure.com/openai/v1/responses/... (fixes the 404s).

If you’d prefer, I can also open a small PR from my fork targeting the gpt-5-pro branch with this change.

---

After applying the change above, everything has been working great with Azure OpenAI.
Thank you very much for the awesome implementation!

[hannesrudolph]

Closing for now as it is way off from the current codebase.