fix(catalog): omit unsupported reasoning.summary on paid xAI Responses

First-party xAI /v1/responses rejects reasoning.summary. Bake
supportsReasoningSummary=false for both xai and xai-oauth so paid
grok-4.5 effort requests send only reasoning.effort, matching SuperGrok.

docs/provider-endpoint-constraints.md

@@ -240,7 +240,8 @@ Reasoning fields are not interchangeable.
 Both the paid API-key provider (`xai` / `XAI_API_KEY`) and SuperGrok OAuth
 (`xai-oauth`) chat over `https://api.x.ai/v1/responses`. Keep these independent:
 
-- omit `reasoning.effort`
+- omit `reasoning.effort` unless the model is on the Grok effort-capable allowlist
+- omit `reasoning.summary` (the host rejects it; do not fall back to `"auto"`)
 - include `reasoning.encrypted_content` on the request
 - replay encrypted reasoning items on later turns
 

packages/ai/CHANGELOG.md

@@ -152,6 +152,10 @@
 ### Fixed
 
 - Fixed an issue where Ollama requests without a user-role message would fail to generate output or silently fail with a misleading error.
+### Fixed
+
+- Stopped treating `XAI_API_KEY` as SuperGrok (`xai-oauth`) sign-in for availability, so paid-key-only setups default to `xai/grok-4.5` instead of the zero-cost SuperGrok catalog path.
+- Omitted unsupported `reasoning.summary` on paid xAI Responses requests (`xai/grok-4.5`), matching SuperGrok, so a thinking level no longer serializes `summary: "auto"`.
 
 ## [17.2.5] - 2026-08-03
 

packages/ai/src/providers/openai-responses.ts

@@ -1285,12 +1285,11 @@ export function buildParams(
 		filterReasoningHistory: options?.filterReasoningHistory,
 		omitReasoningEffort: options?.omitReasoningEffort,
 	});
-	const reasoningSummary =
-		model.provider === "xai-oauth"
-			? options?.reasoning === undefined
-				? undefined
-				: null
-			: options?.reasoningSummary;
+	const reasoningSummary = model.compat.supportsReasoningSummary
+		? options?.reasoningSummary
+		: options?.reasoning === undefined
+			? undefined
+			: null;
 	applyResponsesCompatPolicy(params, reasoningPolicy, {
 		reasoningSummary,
 		forceReasoningOff: options?.forceReasoningOff,

packages/ai/test/xai-oauth-effort-strip.test.ts

@@ -87,6 +87,16 @@ describe("xAI OAuth Responses reasoning payload (regression)", () => {
 		expect(params.include).toContain("reasoning.encrypted_content");
 	});
 
+	test("paid xai/grok-4.5 omits unsupported reasoning summary", () => {
+		const grok45 = getBundledModel<"openai-responses">("xai", "grok-4.5");
+		if (!grok45) throw new Error("xai/grok-4.5 must be in bundled models.json");
+
+		const { params } = buildParams(grok45, singleUserContext, { reasoning: Effort.High }, undefined);
+
+		expect(params.reasoning).toEqual({ effort: "high" });
+		expect(params.include).toContain("reasoning.encrypted_content");
+	});
+
 	test("paid xai/grok-4.5 requests encrypted reasoning content", () => {
 		const grok45 = getBundledModel<"openai-responses">("xai", "grok-4.5");
 		if (!grok45) throw new Error("xai/grok-4.5 must be in bundled models.json");
@@ -117,7 +127,7 @@ describe("xAI OAuth Responses reasoning payload (regression)", () => {
 
 		const { params } = buildParams(grok45, singleUserContext, { reasoning: Effort.Minimal }, undefined);
 
-		expect(params.reasoning).toMatchObject({ effort: "low" });
+		expect(params.reasoning).toEqual({ effort: "low" });
 	});
 
 	test("xai-oauth/grok-4.5 clamps minimal reasoning effort to low", () => {
@@ -126,7 +136,7 @@ describe("xAI OAuth Responses reasoning payload (regression)", () => {
 
 		const { params } = buildParams(grok45, singleUserContext, { reasoning: Effort.Minimal }, undefined);
 
-		expect(params.reasoning).toMatchObject({ effort: "low" });
+		expect(params.reasoning).toEqual({ effort: "low" });
 	});
 
 	test("xai-oauth/grok-4.5 replays encrypted reasoning on the next turn", () => {

packages/catalog/CHANGELOG.md

@@ -142,6 +142,23 @@
 - Fixed dynamic discovery for the `deepseek-v4` model family (such as `deepseek-v4-flash-0731`) under `alibaba-token-plan` missing reasoning configuration and maximum thinking effort.
 - Fixed GitHub Copilot dynamic discovery retaining stale bundled prices for default-context models instead of using the provider's reported default-tier prices.
 
+## [17.2.5] - 2026-08-03
+### Changed
+
+- Switched the paid xAI provider (`xai` / `XAI_API_KEY`) from Chat Completions to the OpenAI Responses API (`POST https://api.x.ai/v1/responses`), matching SuperGrok `xai-oauth`. Prompt-cache affinity (`x-grok-conv-id`), reasoning-effort allowlisting, and encrypted-reasoning replay rules are now shared across both first-party xAI hosts.
+- Changed the paid xAI (`XAI_API_KEY`) default model from `grok-4-fast-non-reasoning` to `grok-4.5`.
+- Changed the SuperGrok (`xai-oauth`) default model from `grok-4.3` to `grok-4.5`.
+- Requested `reasoning.encrypted_content` on first-party xAI Responses calls (`xai` and `xai-oauth`) via the `include` parameter.
+- Replayed xAI encrypted reasoning items on later Responses turns instead of stripping `type: "reasoning"` history.
+
+### Fixed
+
+- Invalidated stale paid-xAI model-cache rows written under Chat Completions so the Responses migration takes effect immediately instead of waiting for TTL expiry.
+- Clamped paid xAI Responses `minimal` reasoning effort to `low` (same wire map as SuperGrok) so `xai/grok-4.5` does not 400.
+- Suppressed presence/frequency penalties and stop sequences on xAI reasoning models so a configured `presencePenalty` does not 400 after the `grok-4.5` default change.
+- Stopped emitting stale `thinking.efforts` dials on paid xAI Responses catalog rows that reject `reasoning.effort` (`grok-code-fast-1`, `grok-build-0.1`, `grok-4.20-0309-reasoning`, and other off-allowlist reasoners).
+- Marked first-party xAI Responses hosts (`xai` and `xai-oauth`) as not supporting `reasoning.summary`, so paid `xai/grok-4.5` effort requests omit the unsupported field instead of sending `summary: "auto"`.
+
 ## [17.2.5] - 2026-08-03
 
 ### Fixed

packages/catalog/src/compat/openai.ts

@@ -713,6 +713,8 @@ export function buildOpenAIResponsesCompat(spec: OpenAIResponsesSpecLike): Resol
 		// Copilot host under a different provider id still clamps.
 		supportsImageDetailOriginal:
 			!isXaiHost && !modelMatchesHost({ provider: spec.provider, baseUrl }, "githubCopilot"),
+		// api.x.ai rejects `reasoning.summary` (SuperGrok and paid key alike).
+		supportsReasoningSummary: !isXaiHost,
 		reasoningEffortMap: isXaiHost ? { ...XAI_RESPONSES_REASONING_EFFORT_MAP } : {},
 		supportsReasoningParams: true,
 		// OpenAI proprietary reasoning models (o-series, gpt-5+) reject explicit
@@ -796,6 +798,7 @@ function pickResponsesOnly(compat: ResolvedOpenAIResponsesCompat): ResponsesOnly
 		strictResponsesPairing: compat.strictResponsesPairing,
 		supportsImageDetailOriginal: compat.supportsImageDetailOriginal,
 		supportsObfuscationOptOut: compat.supportsObfuscationOptOut,
+		supportsReasoningSummary: compat.supportsReasoningSummary,
 		isVercelGatewayHost: compat.isVercelGatewayHost,
 	} satisfies ResponsesOnlyCompat;
 }

packages/catalog/src/types.ts

@@ -719,6 +719,12 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
 	strictResponsesPairing: boolean;
 	supportsImageDetailOriginal: boolean;
 	supportsObfuscationOptOut: boolean;
+	/**
+	 * Whether `reasoning.summary` may be sent. First-party xAI `/v1/responses`
+	 * rejects the field; handlers pass `null` so the wire omits it instead of
+	 * filling `"auto"`.
+	 */
+	supportsReasoningSummary: boolean;
 	streamIdleTimeoutMs?: number;
 	vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
 	/** The model sits behind Vercel AI Gateway's Responses endpoint. */

packages/catalog/test/build.test.ts

@@ -268,6 +268,8 @@ describe("xAI Responses reasoning-effort suppression", () => {
 		expect(oauth.reasoningEffortMap).toEqual({ minimal: "low" });
 		expect(paid.supportsPenaltyAndStopParams).toBe(false);
 		expect(oauth.supportsPenaltyAndStopParams).toBe(false);
+		expect(paid.supportsReasoningSummary).toBe(false);
+		expect(oauth.supportsReasoningSummary).toBe(false);
 	});
 
 	it("keeps penalty and stop params on non-reasoning paid xAI models", () => {