diff --git a/packages/ai/src/providers/gitlab-duo.ts b/packages/ai/src/providers/gitlab-duo.ts
index 834717c09..8ddc8cf3b 100644
--- a/packages/ai/src/providers/gitlab-duo.ts
+++ b/packages/ai/src/providers/gitlab-duo.ts
@@ -267,6 +267,11 @@ export function streamGitLabDuo(
 				...options.headers,
 			};
 
+			const reasoningEffort =
+				options.reasoning === "off"
+					? undefined
+					: (options.reasoning as "minimal" | "low" | "medium" | "high" | "xhigh" | undefined);
+
 			const inner =
 				mapping.provider === "anthropic"
 					? streamAnthropic(
@@ -295,11 +300,11 @@ export function streamGitLabDuo(
 								sessionId: options.sessionId,
 								providerSessionState: options.providerSessionState,
 								onPayload: options.onPayload,
-								thinkingEnabled: Boolean(options.reasoning) && model.reasoning,
-								thinkingBudgetTokens: options.reasoning
-									? (options.thinkingBudgets?.[options.reasoning] ?? ANTHROPIC_THINKING[options.reasoning])
+								thinkingEnabled: Boolean(reasoningEffort) && model.reasoning,
+								thinkingBudgetTokens: reasoningEffort
+									? (options.thinkingBudgets?.[reasoningEffort] ?? ANTHROPIC_THINKING[reasoningEffort])
 									: undefined,
-								reasoning: options.reasoning,
+								reasoning: reasoningEffort,
 								toolChoice: mapAnthropicToolChoice(options.toolChoice),
 							},
 						)
@@ -329,7 +334,7 @@ export function streamGitLabDuo(
 									sessionId: options.sessionId,
 									providerSessionState: options.providerSessionState,
 									onPayload: options.onPayload,
-									reasoning: options.reasoning,
+									reasoning: reasoningEffort,
 									toolChoice: options.toolChoice,
 								} satisfies OpenAIResponsesOptions,
 							)
@@ -358,7 +363,7 @@ export function streamGitLabDuo(
 									sessionId: options.sessionId,
 									providerSessionState: options.providerSessionState,
 									onPayload: options.onPayload,
-									thinkingLevel: options.reasoning,
+									reasoning: reasoningEffort,
 									toolChoice: options.toolChoice,
 								} satisfies OpenAICompletionsOptions,
 							);
diff --git a/packages/ai/src/providers/kimi.ts b/packages/ai/src/providers/kimi.ts
index 07e7e9108..f9729bf79 100644
--- a/packages/ai/src/providers/kimi.ts
+++ b/packages/ai/src/providers/kimi.ts
@@ -62,9 +62,10 @@ export function streamKimi(
 
 				// Calculate thinking budget from reasoning level
 				const reasoning = options?.reasoning;
-				const thinkingEnabled = !!reasoning && model.reasoning;
-				const thinkingBudget = reasoning
-					? (options?.thinkingBudgets?.[reasoning] ?? ANTHROPIC_THINKING[reasoning])
+				const reasoningEffort = reasoning === "off" ? undefined : reasoning;
+				const thinkingEnabled = !!reasoningEffort && model.reasoning;
+				const thinkingBudget = reasoningEffort
+					? (options?.thinkingBudgets?.[reasoningEffort] ?? ANTHROPIC_THINKING[reasoningEffort])
 					: undefined;
 
 				const innerStream = streamAnthropic(anthropicModel, context, {
@@ -89,6 +90,7 @@ export function streamKimi(
 				}
 			} else {
 				// OpenAI format - use original model with Kimi headers
+				const reasoningEffort = options?.reasoning === "off" ? undefined : options?.reasoning;
 				const innerStream = streamOpenAICompletions(model, context, {
 					apiKey: options?.apiKey,
 					temperature: options?.temperature,
@@ -102,7 +104,7 @@ export function streamKimi(
 					headers: mergedHeaders,
 					sessionId: options?.sessionId,
 					onPayload: options?.onPayload,
-					thinkingLevel: options?.reasoning,
+					reasoning: reasoningEffort,
 				});
 
 				for await (const event of innerStream) {
diff --git a/packages/ai/src/providers/synthetic.ts b/packages/ai/src/providers/synthetic.ts
index d6ea05653..3d6dd1d54 100644
--- a/packages/ai/src/providers/synthetic.ts
+++ b/packages/ai/src/providers/synthetic.ts
@@ -59,9 +59,10 @@ export function streamSynthetic(
 
 				// Calculate thinking budget from reasoning level
 				const reasoning = options?.reasoning;
-				const thinkingEnabled = !!reasoning && model.reasoning;
-				const thinkingBudget = reasoning
-					? (options?.thinkingBudgets?.[reasoning] ?? ANTHROPIC_THINKING[reasoning])
+				const reasoningEffort = reasoning === "off" ? undefined : reasoning;
+				const thinkingEnabled = !!reasoningEffort && model.reasoning;
+				const thinkingBudget = reasoningEffort
+					? (options?.thinkingBudgets?.[reasoningEffort] ?? ANTHROPIC_THINKING[reasoningEffort])
 					: undefined;
 
 				const innerStream = streamAnthropic(anthropicModel, context, {
@@ -92,6 +93,7 @@ export function streamSynthetic(
 					headers: mergedHeaders,
 				};
 
+				const reasoningEffort = options?.reasoning === "off" ? undefined : options?.reasoning;
 				const innerStream = streamOpenAICompletions(syntheticModel, context, {
 					apiKey: options?.apiKey,
 					temperature: options?.temperature,
@@ -105,7 +107,7 @@ export function streamSynthetic(
 					headers: mergedHeaders,
 					sessionId: options?.sessionId,
 					onPayload: options?.onPayload,
-					thinkingLevel: options?.reasoning,
+					reasoning: reasoningEffort,
 				});
 
 				for await (const event of innerStream) {
diff --git a/packages/ai/test/stream.test.ts b/packages/ai/test/stream.test.ts
index 94247eff6..fd5180ca6 100644
--- a/packages/ai/test/stream.test.ts
+++ b/packages/ai/test/stream.test.ts
@@ -534,7 +534,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle thinking",
 			async () => {
-				await handleThinking(llm, { thinkingLevel: "high" });
+				await handleThinking(llm, { reasoning: "high" });
 			},
 			{ retry: 2 },
 		);
@@ -542,7 +542,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle multi-turn with thinking and tools",
 			async () => {
-				await multiTurn(llm, { thinkingLevel: "high" });
+				await multiTurn(llm, { reasoning: "high" });
 			},
 			{ retry: 3 },
 		);
@@ -658,7 +658,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle thinking mode",
 			async () => {
-				await handleThinking(llm, { thinkingLevel: "medium" });
+				await handleThinking(llm, { reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
@@ -666,7 +666,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle multi-turn with thinking and tools",
 			async () => {
-				await multiTurn(llm, { thinkingLevel: "medium" });
+				await multiTurn(llm, { reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
@@ -702,7 +702,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle thinking mode",
 			async () => {
-				await handleThinking(llm, { thinkingLevel: "medium" });
+				await handleThinking(llm, { reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
@@ -710,7 +710,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle multi-turn with thinking and tools",
 			async () => {
-				await multiTurn(llm, { thinkingLevel: "medium" });
+				await multiTurn(llm, { reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
@@ -746,7 +746,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle thinking mode",
 			async () => {
-				await handleThinking(llm, { thinkingLevel: "medium" });
+				await handleThinking(llm, { reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
@@ -754,7 +754,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle multi-turn with thinking and tools",
 			async () => {
-				await multiTurn(llm, { thinkingLevel: "medium" });
+				await multiTurn(llm, { reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
@@ -790,7 +790,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle thinking mode",
 			async () => {
-				await handleThinking(llm, { thinkingLevel: "medium" });
+				await handleThinking(llm, { reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
@@ -798,7 +798,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle multi-turn with thinking and tools",
 			async () => {
-				await multiTurn(llm, { thinkingLevel: "medium" });
+				await multiTurn(llm, { reasoning: "medium" });
 			},
 			{ retry: 2 },
 		);
@@ -842,7 +842,7 @@ describe("Generate E2E Tests", () => {
 		it.skip(
 			"should handle thinking mode",
 			async () => {
-				await handleThinking(llm, { thinkingLevel: "medium" });
+				await handleThinking(llm, { reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
@@ -850,7 +850,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle multi-turn with thinking and tools",
 			async () => {
-				await multiTurn(llm, { thinkingLevel: "medium" });
+				await multiTurn(llm, { reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
@@ -886,7 +886,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle thinking mode",
 			async () => {
-				await handleThinking(llm, { thinkingLevel: "medium" });
+				await handleThinking(llm, { reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
@@ -894,7 +894,7 @@ describe("Generate E2E Tests", () => {
 		it(
 			"should handle multi-turn with thinking and tools",
 			async () => {
-				await multiTurn(llm, { thinkingLevel: "medium" });
+				await multiTurn(llm, { reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
@@ -950,7 +950,7 @@ describe("Generate E2E Tests", () => {
 			it(
 				"should handle multi-turn with thinking and tools",
 				async () => {
-					await multiTurn(llm, { thinkingLevel: "medium" });
+					await multiTurn(llm, { reasoning: "medium" });
 				},
 				{ retry: 3 },
 			);
@@ -1076,7 +1076,7 @@ describe("Generate E2E Tests", () => {
 			"should handle thinking",
 			async () => {
 				const thinkingModel = getBundledModel("github-copilot", "gpt-5-mini");
-				await handleThinking(thinkingModel, { apiKey: githubCopilotToken, thinkingLevel: "high" });
+				await handleThinking(thinkingModel, { apiKey: githubCopilotToken, reasoning: "high" });
 			},
 			{ retry: 2 },
 		);
@@ -1085,7 +1085,7 @@ describe("Generate E2E Tests", () => {
 			"should handle multi-turn with thinking and tools",
 			async () => {
 				const thinkingModel = getBundledModel("github-copilot", "gpt-5-mini");
-				await multiTurn(thinkingModel, { apiKey: githubCopilotToken, thinkingLevel: "high" });
+				await multiTurn(thinkingModel, { apiKey: githubCopilotToken, reasoning: "high" });
 			},
 			{ retry: 3 },
 		);
@@ -1318,7 +1318,7 @@ describe("Generate E2E Tests", () => {
 		it.skipIf(!openaiCodexToken)(
 			"should handle thinking",
 			async () => {
-				await handleThinking(llm, { apiKey: openaiCodexToken, thinkingLevel: "high" });
+				await handleThinking(llm, { apiKey: openaiCodexToken, reasoning: "high" });
 			},
 			{ retry: 3 },
 		);
@@ -1490,7 +1490,7 @@ describe("Generate E2E Tests", () => {
 			"should handle thinking mode",
 			async () => {
 				if (!llm) return;
-				await handleThinking(llm, { apiKey: "test", thinkingLevel: "medium" });
+				await handleThinking(llm, { apiKey: "test", reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
@@ -1499,7 +1499,7 @@ describe("Generate E2E Tests", () => {
 			"should handle multi-turn with thinking and tools",
 			async () => {
 				if (!llm) return;
-				await multiTurn(llm, { apiKey: "test", thinkingLevel: "medium" });
+				await multiTurn(llm, { apiKey: "test", reasoning: "medium" });
 			},
 			{ retry: 3 },
 		);
diff --git a/packages/ai/test/xhigh.test.ts b/packages/ai/test/xhigh.test.ts
index 1d3d4955a..5fa795e1a 100644
--- a/packages/ai/test/xhigh.test.ts
+++ b/packages/ai/test/xhigh.test.ts
@@ -21,7 +21,7 @@ describe.skipIf(!e2eApiKey("OPENAI_API_KEY"))("xhigh reasoning", () => {
 		// Note: codex models only support the responses API, not chat completions
 		it("should work with openai-responses", async () => {
 			const model = getBundledModel("openai", "gpt-5.1-codex-max");
-			const s = stream(model, makeContext(), { thinkingLevel: "xhigh" });
+			const s = stream(model, makeContext(), { reasoning: "xhigh" });
 			let hasThinking = false;
 
 			for await (const event of s) {
@@ -40,7 +40,7 @@ describe.skipIf(!e2eApiKey("OPENAI_API_KEY"))("xhigh reasoning", () => {
 	describe("gpt-5-mini (does not support xhigh)", () => {
 		it("should error with openai-responses when using xhigh", async () => {
 			const model = getBundledModel("openai", "gpt-5-mini");
-			const s = stream(model, makeContext(), { thinkingLevel: "xhigh" });
+			const s = stream(model, makeContext(), { reasoning: "xhigh" });
 
 			for await (const _ of s) {
 				// drain events
@@ -56,7 +56,7 @@ describe.skipIf(!e2eApiKey("OPENAI_API_KEY"))("xhigh reasoning", () => {
 				...getBundledModel("openai", "gpt-5-mini"),
 				api: "openai-completions",
 			};
-			const s = stream(model, makeContext(), { thinkingLevel: "xhigh" });
+			const s = stream(model, makeContext(), { reasoning: "xhigh" });
 
 			for await (const _ of s) {
 				// drain events