feat(models): add gpt-image-2 from openai

apps/gateway/src/chat/chat.ts

@@ -3834,7 +3834,7 @@ chat.openapi(completions, async (c) => {
 		}
 	}
 
-	let requestBody: ProviderRequestBody = await prepareRequestBody(
+	let requestBody: ProviderRequestBody | FormData = await prepareRequestBody(
 		usedProvider,
 		upstreamModelName,
 		messages as BaseMessage[],
@@ -3863,6 +3863,7 @@ chat.openapi(completions, async (c) => {
 
 	// Validate effective max_tokens value after prepareRequestBody
 	if (
+		!(requestBody instanceof FormData) &&
 		hasMaxTokens(requestBody) &&
 		requestBody.max_tokens !== undefined &&
 		finalModelInfo
@@ -3892,11 +3893,23 @@ chat.openapi(completions, async (c) => {
 		isImageGeneration &&
 		usedProvider === "xai" &&
 		url &&
+		!(requestBody instanceof FormData) &&
 		("image" in requestBody || "images" in requestBody)
 	) {
 		url = url.replace("/v1/images/generations", "/v1/images/edits");
 	}
 
+	// Switch OpenAI image generation endpoint to /edits when input images are present.
+	// prepareRequestBody returns a FormData (multipart/form-data) only for this edits flow.
+	if (
+		isImageGeneration &&
+		usedProvider === "openai" &&
+		url &&
+		requestBody instanceof FormData
+	) {
+		url = url.replace("/v1/images/generations", "/v1/images/edits");
+	}
+
 	const startTime = Date.now();
 	const failedEnvKeyIndicesByProvider = new Map<string, Set<number>>();
 	const failedTrackedKeyIdsByProvider = new Map<string, Set<string>>();
@@ -7811,7 +7824,9 @@ chat.openapi(completions, async (c) => {
 				requestId,
 				webSearchEnabled: !!webSearchTool,
 			});
-			headers["Content-Type"] = "application/json";
+			if (!(requestBody instanceof FormData)) {
+				headers["Content-Type"] = "application/json";
+			}
 
 			// Add effort beta header for Anthropic if effort parameter is specified
 			if (usedProvider === "anthropic" && effort !== undefined) {
@@ -7841,7 +7856,10 @@ chat.openapi(completions, async (c) => {
 			res = await fetch(url, {
 				method: "POST",
 				headers,
-				body: JSON.stringify(requestBody),
+				body:
+					requestBody instanceof FormData
+						? requestBody
+						: JSON.stringify(requestBody),
 				signal: fetchSignal,
 			});
 		} catch (error) {

apps/gateway/src/chat/tools/parse-provider-response.ts

@@ -526,6 +526,39 @@ export function parseProviderResponse(
 			break;
 		}
 		default: // OpenAI format
+			// Check if this is an OpenAI image generation response (e.g. gpt-image-2)
+			// Format: { created: number, data: [{ b64_json?: string, url?: string, revised_prompt?: string }], usage?: {...} }
+			if (
+				usedProvider === "openai" &&
+				json.data &&
+				Array.isArray(json.data) &&
+				json.data.length > 0 &&
+				(json.data[0]?.b64_json || json.data[0]?.url)
+			) {
+				const imageData = json.data;
+				images = imageData.map((item: any): ImageObject => {
+					let url: string;
+					if (item.b64_json) {
+						url = `data:image/png;base64,${item.b64_json}`;
+					} else {
+						url = item.url;
+					}
+					return {
+						type: "image_url",
+						image_url: { url },
+					};
+				});
+				content = imageLabel;
+				finishReason = "stop";
+				// OpenAI gpt-image models return usage with input/output tokens
+				promptTokens = json.usage?.input_tokens ?? 0;
+				completionTokens = json.usage?.output_tokens ?? 0;
+				cachedTokens = json.usage?.input_tokens_details?.cached_tokens ?? null;
+				totalTokens =
+					json.usage?.total_tokens ??
+					(promptTokens ?? 0) + (completionTokens ?? 0);
+				break;
+			}
 			// Check if this is an xAI Grok Imagine image generation response
 			// Format: { data: [{ url: "..." }] }
 			if (usedProvider === "xai" && json.data && Array.isArray(json.data)) {

apps/gateway/src/chat/tools/resolve-provider-context.ts

@@ -43,7 +43,7 @@ export interface ProviderContext {
 	configIndex: number;
 	envVarName: string | undefined;
 	url: string;
-	requestBody: ProviderRequestBody;
+	requestBody: ProviderRequestBody | FormData;
 	useResponsesApi: boolean;
 	requestCanBeCanceled: boolean;
 	isImageGeneration: boolean;
@@ -365,7 +365,7 @@ export async function resolveProviderContext(
 		providers.find((p) => p.id === usedProvider)?.cancellation === true;
 
 	// --- Request body preparation ---
-	const requestBody: ProviderRequestBody = await prepareRequestBody(
+	const requestBody: ProviderRequestBody | FormData = await prepareRequestBody(
 		usedProvider as Provider,
 		upstreamModelName,
 		options.messages as BaseMessage[],
@@ -394,6 +394,7 @@ export async function resolveProviderContext(
 
 	// Post-validation of max_tokens in request body
 	if (
+		!(requestBody instanceof FormData) &&
 		hasMaxTokens(requestBody) &&
 		requestBody.max_tokens !== undefined &&
 		providerMappingForSelected

apps/gateway/src/chat/tools/transform-response-to-openai.ts

@@ -578,6 +578,54 @@ export function transformResponseToOpenai(
 		case "mistral":
 		case "novita":
 		case "openai": {
+			// Handle OpenAI image generation responses (e.g. gpt-image-2)
+			// Format: { created: number, data: [{ b64_json?: string, url?: string }], usage?: {...} }
+			if (
+				usedProvider === "openai" &&
+				json.data &&
+				Array.isArray(json.data) &&
+				!json.choices &&
+				!json.output
+			) {
+				transformedResponse = {
+					id: `chatcmpl-${Date.now()}`,
+					object: "chat.completion",
+					created: json.created ?? Math.floor(Date.now() / 1000),
+					model: `${usedProvider}/${baseModelName}`,
+					choices: [
+						{
+							index: 0,
+							message: {
+								role: "assistant",
+								content: content,
+								...(images && images.length > 0 && { images }),
+							},
+							finish_reason: finishReason ?? "stop",
+						},
+					],
+					usage: buildUsageObject(
+						promptTokens,
+						completionTokens,
+						totalTokens,
+						reasoningTokens,
+						cachedTokens,
+						costs,
+						showUpgradeMessage,
+						cacheCreationTokens,
+					),
+					metadata: buildMetadata(
+						requestedModel,
+						requestedProvider,
+						baseModelName,
+						usedProvider,
+						usedModel,
+						requestId,
+						routing,
+						usedRegion,
+					),
+				};
+				break;
+			}
 			// Handle OpenAI responses format transformation to chat completions format
 			if (json.output && Array.isArray(json.output)) {
 				// This is from the responses endpoint - transform to chat completions format

apps/playground/src/components/playground/image-page-client.tsx

@@ -152,22 +152,30 @@ export default function ImagePageClient({
 
 	// Keep URL in sync with selected model(s)
 	useEffect(() => {
-		const params = new URLSearchParams(Array.from(searchParams.entries()));
+		// Read current URL params directly to avoid stale searchParams closure
+		// and to prevent an infinite loop where router.replace produces a new
+		// searchParams reference that re-triggers this effect (each such cycle
+		// causes Next.js to refetch the RSC, re-hitting /orgs forever).
+		const currentParams = new URLSearchParams(window.location.search);
 		if (comparisonMode) {
-			params.set("model", selectedModels.join(","));
-			params.set("compare", "1");
+			currentParams.set("model", selectedModels.join(","));
+			currentParams.set("compare", "1");
 		} else {
 			const primary = selectedModels[0];
 			if (primary) {
-				params.set("model", primary);
+				currentParams.set("model", primary);
 			} else {
-				params.delete("model");
+				currentParams.delete("model");
 			}
-			params.delete("compare");
+			currentParams.delete("compare");
 		}
-		const qs = params.toString();
-		router.replace(qs ? `?${qs}` : "");
-	}, [comparisonMode, router, searchParams, selectedModels]);
+		const qs = currentParams.toString();
+		const nextUrl = `${pathname}${qs ? `?${qs}` : ""}`;
+		const currentUrl = `${window.location.pathname}${window.location.search}`;
+		if (nextUrl !== currentUrl) {
+			router.replace(nextUrl, { scroll: false });
+		}
+	}, [comparisonMode, pathname, router, selectedModels]);
 
 	// Reset imageSize when model changes, clear input images when switching away from edit model
 	useEffect(() => {

apps/playground/src/components/playground/video-page-client.tsx

@@ -323,22 +323,30 @@ export default function VideoPageClient({
 
 	// Keep URL in sync with selected model(s)
 	useEffect(() => {
-		const params = new URLSearchParams(Array.from(searchParams.entries()));
+		// Read current URL params directly to avoid stale searchParams closure
+		// and to prevent an infinite loop where router.replace produces a new
+		// searchParams reference that re-triggers this effect (each such cycle
+		// causes Next.js to refetch the RSC, re-hitting /orgs forever).
+		const currentParams = new URLSearchParams(window.location.search);
 		if (comparisonMode) {
-			params.set("model", selectedModels.join(","));
-			params.set("compare", "1");
+			currentParams.set("model", selectedModels.join(","));
+			currentParams.set("compare", "1");
 		} else {
 			const primary = selectedModels[0];
 			if (primary) {
-				params.set("model", primary);
+				currentParams.set("model", primary);
 			} else {
-				params.delete("model");
+				currentParams.delete("model");
 			}
-			params.delete("compare");
+			currentParams.delete("compare");
 		}
-		const qs = params.toString();
-		router.replace(qs ? `?${qs}` : "");
-	}, [comparisonMode, router, searchParams, selectedModels]);
+		const qs = currentParams.toString();
+		const nextUrl = `${pathname}${qs ? `?${qs}` : ""}`;
+		const currentUrl = `${window.location.pathname}${window.location.search}`;
+		if (nextUrl !== currentUrl) {
+			router.replace(nextUrl, { scroll: false });
+		}
+	}, [comparisonMode, pathname, router, selectedModels]);
 
 	const getModelName = useCallback(
 		(modelId: string) => {

packages/actions/src/get-provider-endpoint.ts

@@ -403,6 +403,9 @@ export function getProviderEndpoint(
 			}
 		}
 		case "openai": {
+			if (imageGenerations) {
+				return `${url}/v1/images/generations`;
+			}
 			// Use responses endpoint for models that support responses API
 			if (model) {
 				// Look up by model ID first, then fall back to provider modelName