diff --git a/apps/gateway/src/chat/tools/extract-token-usage.spec.ts b/apps/gateway/src/chat/tools/extract-token-usage.spec.ts
index 7883c6009..dd7d81cad 100644
--- a/apps/gateway/src/chat/tools/extract-token-usage.spec.ts
+++ b/apps/gateway/src/chat/tools/extract-token-usage.spec.ts
@@ -26,6 +26,28 @@ describe("extractTokenUsage", () => {
 			expect(result.totalTokens).toBe(350);
 		});
 
+		it("extracts cache creation tokens from cacheDetails by TTL", () => {
+			const data = {
+				usage: {
+					inputTokens: 100,
+					cacheReadInputTokens: 0,
+					cacheWriteInputTokens: 1000,
+					cacheDetails: [
+						{ ttl: "1h", inputTokens: 700 },
+						{ ttl: "5m", inputTokens: 300 },
+					],
+					outputTokens: 200,
+					totalTokens: 1300,
+				},
+			};
+
+			const result = extractTokenUsage(data, "aws-bedrock");
+
+			expect(result.cacheCreationTokens).toBe(1000);
+			expect(result.cacheCreation5mTokens).toBe(300);
+			expect(result.cacheCreation1hTokens).toBe(700);
+		});
+
 		it("returns cachedTokens with correct value when cacheReadInputTokens > 0", () => {
 			const data = {
 				usage: {
diff --git a/apps/gateway/src/chat/tools/extract-token-usage.ts b/apps/gateway/src/chat/tools/extract-token-usage.ts
index 4b5ec12d8..ffd79875b 100644
--- a/apps/gateway/src/chat/tools/extract-token-usage.ts
+++ b/apps/gateway/src/chat/tools/extract-token-usage.ts
@@ -31,6 +31,31 @@ export function adjustGoogleCandidateTokens(
 	return candidatesTokenCount;
 }
 
+export function extractBedrockCacheCreationDetails(usage: any): {
+	cacheCreation5mTokens: number | null;
+	cacheCreation1hTokens: number | null;
+} {
+	let fiveMinuteTokens = 0;
+	let oneHourTokens = 0;
+
+	const cacheDetails = Array.isArray(usage?.cacheDetails)
+		? usage.cacheDetails
+		: [];
+	for (const detail of cacheDetails) {
+		const inputTokens = detail?.inputTokens ?? 0;
+		if (detail?.ttl === "1h") {
+			oneHourTokens += inputTokens;
+		} else if (detail?.ttl === "5m") {
+			fiveMinuteTokens += inputTokens;
+		}
+	}
+
+	return {
+		cacheCreation5mTokens: fiveMinuteTokens > 0 ? fiveMinuteTokens : null,
+		cacheCreation1hTokens: oneHourTokens > 0 ? oneHourTokens : null,
+	};
+}
+
 /**
  * Extracts token usage information from streaming data based on provider format
  */
@@ -106,6 +131,7 @@ export function extractTokenUsage(
 				const inputTokens = data.usage.inputTokens ?? 0;
 				const cacheReadTokens = data.usage.cacheReadInputTokens ?? 0;
 				const cacheWriteTokens = data.usage.cacheWriteInputTokens ?? 0;
+				const cacheDetails = extractBedrockCacheCreationDetails(data.usage);
 
 				// Total prompt tokens = regular input + cache read + cache write
 				promptTokens = inputTokens + cacheReadTokens + cacheWriteTokens;
@@ -113,6 +139,8 @@ export function extractTokenUsage(
 				// Cached tokens are the tokens read from cache (discount applies to these)
 				cachedTokens = cacheReadTokens;
 				cacheCreationTokens = cacheWriteTokens;
+				cacheCreation5mTokens = cacheDetails.cacheCreation5mTokens;
+				cacheCreation1hTokens = cacheDetails.cacheCreation1hTokens;
 				totalTokens = data.usage.totalTokens ?? null;
 			}
 			break;
diff --git a/apps/gateway/src/chat/tools/parse-provider-response.spec.ts b/apps/gateway/src/chat/tools/parse-provider-response.spec.ts
index 80cdbef83..68edda9e6 100644
--- a/apps/gateway/src/chat/tools/parse-provider-response.spec.ts
+++ b/apps/gateway/src/chat/tools/parse-provider-response.spec.ts
@@ -84,6 +84,39 @@ describe("parseProviderResponse", () => {
 			expect(result.promptTokens).toBe(150); // 100 + 0 + 50
 		});
 
+		it("extracts cache creation tokens from cacheDetails by TTL", () => {
+			const json = {
+				output: {
+					message: {
+						content: [{ text: "Hello" }],
+						role: "assistant",
+					},
+				},
+				stopReason: "end_turn",
+				usage: {
+					inputTokens: 100,
+					cacheReadInputTokens: 0,
+					cacheWriteInputTokens: 1000,
+					cacheDetails: [
+						{ ttl: "1h", inputTokens: 700 },
+						{ ttl: "5m", inputTokens: 300 },
+					],
+					outputTokens: 200,
+					totalTokens: 1300,
+				},
+			};
+
+			const result = parseProviderResponse(
+				"aws-bedrock",
+				"anthropic.claude-sonnet-4-5-20250929-v1:0",
+				json,
+			);
+
+			expect(result.cacheCreationTokens).toBe(1000);
+			expect(result.cacheCreation5mTokens).toBe(300);
+			expect(result.cacheCreation1hTokens).toBe(700);
+		});
+
 		it("returns cachedTokens with correct value when cacheReadInputTokens > 0", () => {
 			const json = {
 				output: {
diff --git a/apps/gateway/src/chat/tools/parse-provider-response.ts b/apps/gateway/src/chat/tools/parse-provider-response.ts
index 5b7793f05..2303d6f9a 100644
--- a/apps/gateway/src/chat/tools/parse-provider-response.ts
+++ b/apps/gateway/src/chat/tools/parse-provider-response.ts
@@ -2,7 +2,10 @@ import { redisClient } from "@llmgateway/cache";
 import { logger } from "@llmgateway/logger";
 
 import { estimateTokens } from "./estimate-tokens.js";
-import { adjustGoogleCandidateTokens } from "./extract-token-usage.js";
+import {
+	adjustGoogleCandidateTokens,
+	extractBedrockCacheCreationDetails,
+} from "./extract-token-usage.js";
 import {
 	extractReasoningDetailsText,
 	splitReasoningFromTaggedContent,
@@ -104,6 +107,7 @@ export function parseProviderResponse(
 				const inputTokens = json.usage.inputTokens ?? 0;
 				const cacheReadTokens = json.usage.cacheReadInputTokens ?? 0;
 				const cacheWriteTokens = json.usage.cacheWriteInputTokens ?? 0;
+				const cacheDetails = extractBedrockCacheCreationDetails(json.usage);
 
 				// Total prompt tokens = regular input + cache read + cache write
 				promptTokens = inputTokens + cacheReadTokens + cacheWriteTokens;
@@ -112,6 +116,8 @@ export function parseProviderResponse(
 				// Cached tokens are the tokens read from cache (discount applies to these)
 				cachedTokens = cacheReadTokens;
 				cacheCreationTokens = cacheWriteTokens;
+				cacheCreation5mTokens = cacheDetails.cacheCreation5mTokens;
+				cacheCreation1hTokens = cacheDetails.cacheCreation1hTokens;
 			}
 
 			// Extract tool calls if present
diff --git a/apps/gateway/src/chat/tools/transform-streaming-to-openai.spec.ts b/apps/gateway/src/chat/tools/transform-streaming-to-openai.spec.ts
index 9805ab18a..d65da50ab 100644
--- a/apps/gateway/src/chat/tools/transform-streaming-to-openai.spec.ts
+++ b/apps/gateway/src/chat/tools/transform-streaming-to-openai.spec.ts
@@ -158,6 +158,50 @@ describe("transformStreamingToOpenai", () => {
 		expect(warn).not.toHaveBeenCalled();
 	});
 
+	it("maps AWS Bedrock metadata cache creation details", () => {
+		warn.mockClear();
+
+		const result = transformStreamingToOpenai(
+			"aws-bedrock",
+			"anthropic.claude-sonnet-4-5-20250929-v1:0",
+			{
+				__aws_event_type: "metadata",
+				usage: {
+					inputTokens: 10,
+					cacheReadInputTokens: 0,
+					cacheWriteInputTokens: 1000,
+					cacheDetails: [
+						{ ttl: "1h", inputTokens: 700 },
+						{ ttl: "5m", inputTokens: 300 },
+					],
+					outputTokens: 1,
+					totalTokens: 1011,
+				},
+			},
+			[],
+		);
+
+		expect(result).toMatchObject({
+			object: "chat.completion.chunk",
+			model: "anthropic.claude-sonnet-4-5-20250929-v1:0",
+			usage: {
+				prompt_tokens: 1010,
+				completion_tokens: 1,
+				total_tokens: 1011,
+				prompt_tokens_details: {
+					cached_tokens: 0,
+					cache_write_tokens: 1000,
+					cache_creation_tokens: 1000,
+					cache_creation: {
+						ephemeral_5m_input_tokens: 300,
+						ephemeral_1h_input_tokens: 700,
+					},
+				},
+			},
+		});
+		expect(warn).not.toHaveBeenCalled();
+	});
+
 	it("treats non-text AWS Bedrock contentBlockDelta members as handled", () => {
 		warn.mockClear();
 
diff --git a/apps/gateway/src/chat/tools/transform-streaming-to-openai.ts b/apps/gateway/src/chat/tools/transform-streaming-to-openai.ts
index 231b2b2a6..13ccc5d21 100644
--- a/apps/gateway/src/chat/tools/transform-streaming-to-openai.ts
+++ b/apps/gateway/src/chat/tools/transform-streaming-to-openai.ts
@@ -3,7 +3,10 @@ import { logger } from "@llmgateway/logger";
 
 import { calculatePromptTokensFromMessages } from "./calculate-prompt-tokens.js";
 import { extractImages } from "./extract-images.js";
-import { adjustGoogleCandidateTokens } from "./extract-token-usage.js";
+import {
+	adjustGoogleCandidateTokens,
+	extractBedrockCacheCreationDetails,
+} from "./extract-token-usage.js";
 import { mapFinishReasonToOpenai } from "./map-finish-reason-to-openai.js";
 import { transformOpenaiStreaming } from "./transform-openai-streaming.js";
 
@@ -1224,7 +1227,11 @@ export function transformStreamingToOpenai(
 				const inputTokens = data.usage.inputTokens ?? 0;
 				const cacheReadTokens = data.usage.cacheReadInputTokens ?? 0;
 				const cacheWriteTokens = data.usage.cacheWriteInputTokens ?? 0;
+				const cacheDetails = extractBedrockCacheCreationDetails(data.usage);
 				const promptTokens = inputTokens + cacheReadTokens + cacheWriteTokens;
+				const hasCacheCreationDetails =
+					cacheDetails.cacheCreation5mTokens !== null ||
+					cacheDetails.cacheCreation1hTokens !== null;
 
 				transformedData = {
 					id: `chatcmpl-${Date.now()}`,
@@ -1242,9 +1249,27 @@ export function transformStreamingToOpenai(
 						prompt_tokens: promptTokens,
 						completion_tokens: data.usage.outputTokens ?? 0,
 						total_tokens: data.usage.totalTokens ?? 0,
-						...(cacheReadTokens > 0 && {
+						...((cacheReadTokens > 0 || cacheWriteTokens > 0) && {
 							prompt_tokens_details: {
 								cached_tokens: cacheReadTokens,
+								...(cacheWriteTokens > 0 && {
+									cache_write_tokens: cacheWriteTokens,
+									cache_creation_tokens: cacheWriteTokens,
+								}),
+								...(cacheWriteTokens > 0 &&
+									hasCacheCreationDetails && {
+										cache_creation: {
+											ephemeral_5m_input_tokens:
+												cacheDetails.cacheCreation5mTokens ??
+												Math.max(
+													0,
+													cacheWriteTokens -
+														(cacheDetails.cacheCreation1hTokens ?? 0),
+												),
+											ephemeral_1h_input_tokens:
+												cacheDetails.cacheCreation1hTokens ?? 0,
+										},
+									}),
 							},
 						}),
 					},
diff --git a/apps/gateway/src/lib/anthropic-pricing.spec.ts b/apps/gateway/src/lib/anthropic-pricing.spec.ts
index 5284f9ad6..330d6369b 100644
--- a/apps/gateway/src/lib/anthropic-pricing.spec.ts
+++ b/apps/gateway/src/lib/anthropic-pricing.spec.ts
@@ -137,3 +137,91 @@ describe("Anthropic model pricing", () => {
 		},
 	);
 });
+
+describe("AWS Bedrock Anthropic model pricing", () => {
+	const bedrockProviderEntries = models.flatMap((model) =>
+		model.family === "anthropic"
+			? model.providers
+					.filter((provider) => provider.providerId === "aws-bedrock")
+					.map((provider) => ({
+						modelId: model.id,
+						provider: provider as ProviderModelMapping,
+					}))
+			: [],
+	);
+
+	it("has at least one AWS Bedrock Anthropic provider mapping to validate", () => {
+		expect(bedrockProviderEntries.length).toBeGreaterThan(0);
+	});
+
+	it.each(bedrockProviderEntries)(
+		"$modelId defines cacheWriteInputPrice whenever cachedInputPrice is set",
+		({ provider }) => {
+			if (provider.cachedInputPrice === undefined) {
+				return;
+			}
+			expect(
+				provider.cacheWriteInputPrice,
+				`${provider.modelName}: cachedInputPrice is set but cacheWriteInputPrice is missing`,
+			).toBeDefined();
+		},
+	);
+
+	const ONE_HOUR_BEDROCK_PREFIXES = [
+		"anthropic.claude-opus-4-5",
+		"anthropic.claude-opus-4-6",
+		"anthropic.claude-opus-4-7",
+		"anthropic.claude-haiku-4-5",
+		"anthropic.claude-sonnet-4-5",
+		"anthropic.claude-sonnet-4-6",
+	];
+	const supportsBedrock1h = (modelName: string) =>
+		ONE_HOUR_BEDROCK_PREFIXES.some((prefix) => modelName.startsWith(prefix));
+
+	it.each(bedrockProviderEntries)(
+		"$modelId only sets cacheWriteInputPrice1h on bedrock models that support 1h TTL",
+		({ provider }) => {
+			if (provider.cacheWriteInputPrice1h === undefined) {
+				return;
+			}
+			expect(
+				supportsBedrock1h(provider.modelName),
+				`${provider.modelName}: cacheWriteInputPrice1h is set but bedrock does not document 1h TTL support for this model`,
+			).toBe(true);
+		},
+	);
+
+	it.each(bedrockProviderEntries)(
+		"$modelId cache prices follow the standard 1.25x/2x/0.1x ratios",
+		({ provider }) => {
+			if (provider.inputPrice === undefined) {
+				return;
+			}
+			const base = provider.inputPrice;
+			if (provider.cacheWriteInputPrice !== undefined) {
+				assertRatio(
+					provider.modelName,
+					"cacheWriteInputPrice (5m)",
+					provider.cacheWriteInputPrice,
+					base * FIVE_MIN_WRITE_MULTIPLIER,
+				);
+			}
+			if (provider.cacheWriteInputPrice1h !== undefined) {
+				assertRatio(
+					provider.modelName,
+					"cacheWriteInputPrice1h",
+					provider.cacheWriteInputPrice1h,
+					base * ONE_HOUR_WRITE_MULTIPLIER,
+				);
+			}
+			if (provider.cachedInputPrice !== undefined) {
+				assertRatio(
+					provider.modelName,
+					"cachedInputPrice",
+					provider.cachedInputPrice,
+					base * CACHE_READ_MULTIPLIER,
+				);
+			}
+		},
+	);
+});
diff --git a/apps/gateway/src/lib/costs.spec.ts b/apps/gateway/src/lib/costs.spec.ts
index bfab453a3..ada191320 100644
--- a/apps/gateway/src/lib/costs.spec.ts
+++ b/apps/gateway/src/lib/costs.spec.ts
@@ -200,6 +200,42 @@ describe("calculateCosts", () => {
 		expect(result.cacheWriteInputCost).toBeCloseTo(1000 * (3.75 / 1e6));
 	});
 
+	it("should calculate AWS Bedrock Claude cache write costs", async () => {
+		// Bedrock Claude Haiku 4.5 input is 1.0/1M; 5m write 1.25/1M; 1h write 2.0/1M.
+		const result = await calculateCosts(
+			"claude-haiku-4-5",
+			"aws-bedrock",
+			1004,
+			50,
+			0,
+			undefined,
+			null,
+			0,
+			undefined,
+			0,
+			null,
+			null,
+			undefined,
+			{
+				cacheWriteTokens: 1000,
+				cacheWrite1hTokens: 700,
+			},
+		);
+
+		const discountMultiplier = 0.8;
+		expect(result.inputCost).toBeCloseTo(4 * (1.0 / 1e6) * discountMultiplier);
+		expect(result.outputCost).toBeCloseTo(
+			50 * (5.0 / 1e6) * discountMultiplier,
+		);
+		const fiveMinuteCacheWriteCost = 300 * (1.25 / 1e6);
+		const oneHourCacheWriteCost = 700 * (2.0 / 1e6);
+		expect(result.cacheWriteInputCost).toBeCloseTo(
+			(fiveMinuteCacheWriteCost + oneHourCacheWriteCost) * discountMultiplier,
+		);
+		expect(result.discount).toBeCloseTo(0.2);
+		expect(result.cacheWriteTokens).toBe(1000);
+	});
+
 	it("should calculate costs with cached tokens for Anthropic (subsequent request - cache read)", async () => {
 		// For Anthropic subsequent request: 4 non-cached + 1659 cache read = 1663 total tokens, 1659 cache reads
 		const result = await calculateCosts(
diff --git a/apps/gateway/src/native-anthropic-cache.e2e.ts b/apps/gateway/src/native-anthropic-cache.e2e.ts
index 8e24cc016..9efb3c1ee 100644
--- a/apps/gateway/src/native-anthropic-cache.e2e.ts
+++ b/apps/gateway/src/native-anthropic-cache.e2e.ts
@@ -588,4 +588,83 @@ describe("e2e native /v1/messages cache", getConcurrentTestOptions(), () => {
 			}
 		},
 	);
+
+	// 1h cache TTL via Bedrock /v1/chat/completions: opts into Bedrock's 1h
+	// cache write rate (2x base) on a model that supports it (Haiku 4.5) and
+	// asserts the gateway forwards ttl:"1h" to the Converse API cachePoint and
+	// surfaces the response breakdown
+	// (prompt_tokens_details.cache_creation.ephemeral_1h_input_tokens) so SDK
+	// clients can attribute spend across rates.
+	(hasBedrockKey ? test : test.skip)(
+		"openai-compat /v1/chat/completions forwards 1h ttl and surfaces cache_creation breakdown for bedrock",
+		getTestOptions(),
+		async () => {
+			const longText = buildLongSystemPrompt();
+			const body = {
+				model: "aws-bedrock/claude-sonnet-4-6",
+				messages: [
+					{
+						role: "system",
+						content: [
+							{
+								type: "text" as const,
+								text: longText,
+								cache_control: {
+									type: "ephemeral" as const,
+									ttl: "1h" as const,
+								},
+							},
+						],
+					},
+					{ role: "user", content: "Just reply OK." },
+				],
+			};
+
+			const send = async () => {
+				const requestId = generateTestRequestId();
+				const res = await app.request("/v1/chat/completions", {
+					method: "POST",
+					headers: {
+						"Content-Type": "application/json",
+						"x-request-id": requestId,
+						Authorization: `Bearer real-token`,
+					},
+					body: JSON.stringify(body),
+				});
+				const json = await res.json();
+				if (logMode) {
+					console.log(
+						"openai-compat bedrock 1h",
+						requestId,
+						"status",
+						res.status,
+						"usage",
+						JSON.stringify(json.usage),
+					);
+				}
+				return { status: res.status, json };
+			};
+
+			const first = await send();
+			expect(first.status).toBe(200);
+			// On the priming call Bedrock should write to the 1h cache. If
+			// prepare-request-body strips ttl, the write falls back to 5m and
+			// ephemeral_1h_input_tokens will be 0 / absent.
+			const cacheWriteTokens =
+				first.json.usage?.prompt_tokens_details?.cache_creation_tokens ??
+				first.json.usage?.prompt_tokens_details?.cache_write_tokens ??
+				0;
+			if (cacheWriteTokens > 0) {
+				const breakdown =
+					first.json.usage?.prompt_tokens_details?.cache_creation;
+				expect(breakdown).toBeDefined();
+				expect(breakdown.ephemeral_1h_input_tokens).toBeGreaterThan(0);
+				expect(breakdown.ephemeral_5m_input_tokens).toBe(0);
+				expect(
+					breakdown.ephemeral_5m_input_tokens +
+						breakdown.ephemeral_1h_input_tokens,
+				).toBe(cacheWriteTokens);
+			}
+		},
+	);
 });
diff --git a/packages/actions/src/prepare-request-body.spec.ts b/packages/actions/src/prepare-request-body.spec.ts
index cd7e4585d..a9f1699c5 100644
--- a/packages/actions/src/prepare-request-body.spec.ts
+++ b/packages/actions/src/prepare-request-body.spec.ts
@@ -787,6 +787,106 @@ describe("prepareRequestBody - MiniMax", () => {
 });
 
 describe("prepareRequestBody - AWS Bedrock", () => {
+	test("should preserve explicit cache_control ttl as Bedrock cachePoint ttl", async () => {
+		const requestBody = (await prepareRequestBody(
+			"aws-bedrock",
+			"anthropic.claude-sonnet-4-5-20250929-v1:0",
+			[
+				{
+					role: "system",
+					content: [
+						{
+							type: "text",
+							text: "Cache this system prompt.",
+							cache_control: { type: "ephemeral", ttl: "1h" },
+						},
+					],
+				},
+				{
+					role: "user",
+					content: [
+						{
+							type: "text",
+							text: "What should I do next?",
+							cache_control: { type: "ephemeral", ttl: "5m" },
+						},
+					],
+				},
+			],
+			false,
+			undefined,
+			undefined,
+			undefined,
+			undefined,
+			undefined,
+			undefined,
+			undefined,
+			undefined,
+			undefined,
+			false,
+			false,
+		)) as any;
+
+		expect(requestBody.system).toEqual([
+			{ text: "Cache this system prompt." },
+			{ cachePoint: { type: "default", ttl: "1h" } },
+		]);
+		expect(requestBody.messages[0].content).toEqual([
+			{ text: "What should I do next?" },
+			{ cachePoint: { type: "default", ttl: "5m" } },
+		]);
+	});
+
+	test("should drop ttl:1h on bedrock models that do not support 1h TTL", async () => {
+		const requestBody = (await prepareRequestBody(
+			"aws-bedrock",
+			"anthropic.claude-3-7-sonnet-20250219-v1:0",
+			[
+				{
+					role: "system",
+					content: [
+						{
+							type: "text",
+							text: "Cache this system prompt.",
+							cache_control: { type: "ephemeral", ttl: "1h" },
+						},
+					],
+				},
+				{
+					role: "user",
+					content: [
+						{
+							type: "text",
+							text: "What should I do next?",
+							cache_control: { type: "ephemeral", ttl: "1h" },
+						},
+					],
+				},
+			],
+			false,
+			undefined,
+			undefined,
+			undefined,
+			undefined,
+			undefined,
+			undefined,
+			undefined,
+			undefined,
+			undefined,
+			false,
+			false,
+		)) as any;
+
+		expect(requestBody.system).toEqual([
+			{ text: "Cache this system prompt." },
+			{ cachePoint: { type: "default" } },
+		]);
+		expect(requestBody.messages[0].content).toEqual([
+			{ text: "What should I do next?" },
+			{ cachePoint: { type: "default" } },
+		]);
+	});
+
 	test("should sanitize complex tool schemas for Bedrock Converse", async () => {
 		const requestBody = (await prepareRequestBody(
 			"aws-bedrock",
diff --git a/packages/actions/src/prepare-request-body.ts b/packages/actions/src/prepare-request-body.ts
index 6276526c9..169c2013c 100644
--- a/packages/actions/src/prepare-request-body.ts
+++ b/packages/actions/src/prepare-request-body.ts
@@ -1624,6 +1624,9 @@ export async function prepareRequestBody(
 			// Track cache control usage (max 4 blocks per Anthropic/Bedrock limit)
 			let bedrockCacheControlCount = 0;
 			const bedrockMaxCacheControlBlocks = 4;
+			interface BedrockCachePoint {
+				cachePoint: { type: "default"; ttl?: "5m" | "1h" };
+			}
 
 			// Get the minCacheableTokens from the model definition (default to 1024 if not specified)
 			const bedrockProviderMapping = modelDef?.providers.find(
@@ -1634,6 +1637,25 @@ export async function prepareRequestBody(
 			// Approximate 4 characters per token
 			const bedrockMinCacheableChars = bedrockMinCacheableTokens * 4;
 
+			// AWS Bedrock supports 1h TTL only on Claude Opus/Haiku/Sonnet 4.5+. For
+			// other models, forwarding ttl:"1h" causes Bedrock to reject the request.
+			// Use cacheWriteInputPrice1h on the model definition as the source of
+			// truth and silently downgrade unsupported 1h hints to the default 5m.
+			const bedrockSupports1hTtl =
+				bedrockProviderMapping?.cacheWriteInputPrice1h !== undefined;
+			const createBedrockCachePoint = (
+				ttl?: "5m" | "1h",
+			): BedrockCachePoint => {
+				const effectiveTtl =
+					ttl === "1h" && !bedrockSupports1hTtl ? undefined : ttl;
+				return {
+					cachePoint: {
+						type: "default",
+						...(effectiveTtl && { ttl: effectiveTtl }),
+					},
+				};
+			};
+
 			// Extract system messages for Bedrock's system field (required for prompt caching)
 			const bedrockSystemMessages = processedMessages.filter(
 				(m) => m.role === "system",
@@ -1649,13 +1671,12 @@ export async function prepareRequestBody(
 			// cachePoint, and fall back to a length heuristic when nothing was
 			// explicitly opted in.
 			if (bedrockSystemMessages.length > 0) {
-				const systemContent: Array<
-					{ text: string } | { cachePoint: { type: "default" } }
-				> = [];
+				const systemContent: Array<{ text: string } | BedrockCachePoint> = [];
 
 				const collectedBedrockBlocks: Array<{
 					text: string;
 					hasExplicitCacheControl: boolean;
+					ttl?: "5m" | "1h";
 				}> = [];
 				for (const sysMsg of bedrockSystemMessages) {
 					if (typeof sysMsg.content === "string") {
@@ -1671,6 +1692,7 @@ export async function prepareRequestBody(
 								collectedBedrockBlocks.push({
 									text: part.text,
 									hasExplicitCacheControl: !!part.cache_control,
+									ttl: part.cache_control?.ttl,
 								});
 							}
 						}
@@ -1687,7 +1709,7 @@ export async function prepareRequestBody(
 					if (block.hasExplicitCacheControl) {
 						if (bedrockCacheControlCount < bedrockMaxCacheControlBlocks) {
 							bedrockCacheControlCount++;
-							systemContent.push({ cachePoint: { type: "default" } });
+							systemContent.push(createBedrockCachePoint(block.ttl));
 						}
 						continue;
 					}
@@ -1699,7 +1721,7 @@ export async function prepareRequestBody(
 
 					if (shouldHeuristicCache) {
 						bedrockCacheControlCount++;
-						systemContent.push({ cachePoint: { type: "default" } });
+						systemContent.push(createBedrockCachePoint());
 					}
 				}
 
@@ -1800,9 +1822,7 @@ export async function prepareRequestBody(
 
 						if (shouldCache) {
 							bedrockCacheControlCount++;
-							bedrockMessage.content.push({
-								cachePoint: { type: "default" },
-							});
+							bedrockMessage.content.push(createBedrockCachePoint());
 						}
 					}
 				} else if (Array.isArray(msg.content)) {
@@ -1818,9 +1838,9 @@ export async function prepareRequestBody(
 								if (part.cache_control) {
 									if (bedrockCacheControlCount < bedrockMaxCacheControlBlocks) {
 										bedrockCacheControlCount++;
-										bedrockMessage.content.push({
-											cachePoint: { type: "default" },
-										});
+										bedrockMessage.content.push(
+											createBedrockCachePoint(part.cache_control.ttl),
+										);
 									}
 								} else {
 									// Add cachePoint as separate block for long text parts
@@ -1831,9 +1851,7 @@ export async function prepareRequestBody(
 
 									if (shouldCache) {
 										bedrockCacheControlCount++;
-										bedrockMessage.content.push({
-											cachePoint: { type: "default" },
-										});
+										bedrockMessage.content.push(createBedrockCachePoint());
 									}
 								}
 							}
@@ -1878,9 +1896,7 @@ export async function prepareRequestBody(
 							boundaryMsg.content[boundaryMsg.content.length - 1];
 						// Only add if the last block isn't already a cachePoint.
 						if (!lastBlock.cachePoint) {
-							boundaryMsg.content.push({
-								cachePoint: { type: "default" },
-							});
+							boundaryMsg.content.push(createBedrockCachePoint());
 							bedrockCacheControlCount++;
 						}
 					}
diff --git a/packages/models/src/models/anthropic.ts b/packages/models/src/models/anthropic.ts
index a39d94358..324c8e398 100644
--- a/packages/models/src/models/anthropic.ts
+++ b/packages/models/src/models/anthropic.ts
@@ -40,6 +40,7 @@ export const anthropicModels = [
 				inputPrice: 3.0 / 1e6,
 				outputPrice: 15.0 / 1e6,
 				cachedInputPrice: 0.3 / 1e6,
+				cacheWriteInputPrice: 3.75 / 1e6,
 				minCacheableTokens: 1024,
 				requestPrice: 0,
 				discount: 0.2,
@@ -154,6 +155,7 @@ export const anthropicModels = [
 				inputPrice: 3.0 / 1e6,
 				outputPrice: 15.0 / 1e6,
 				cachedInputPrice: 0.3 / 1e6,
+				cacheWriteInputPrice: 3.75 / 1e6,
 				minCacheableTokens: 1024,
 				requestPrice: 0,
 				discount: 0.2,
@@ -225,6 +227,7 @@ export const anthropicModels = [
 				inputPrice: 3.0 / 1e6,
 				outputPrice: 15.0 / 1e6,
 				cachedInputPrice: 0.3 / 1e6,
+				cacheWriteInputPrice: 3.75 / 1e6,
 				minCacheableTokens: 1024,
 				requestPrice: 0,
 				discount: 0.2,
@@ -276,6 +279,8 @@ export const anthropicModels = [
 				inputPrice: 3.0 / 1e6,
 				outputPrice: 15.0 / 1e6,
 				cachedInputPrice: 0.3 / 1e6,
+				cacheWriteInputPrice: 3.75 / 1e6,
+				cacheWriteInputPrice1h: 6.0 / 1e6,
 				minCacheableTokens: 1024,
 				requestPrice: 0,
 				discount: 0.2,
@@ -327,6 +332,8 @@ export const anthropicModels = [
 				inputPrice: 3.0 / 1e6,
 				outputPrice: 15.0 / 1e6,
 				cachedInputPrice: 0.3 / 1e6,
+				cacheWriteInputPrice: 3.75 / 1e6,
+				cacheWriteInputPrice1h: 6.0 / 1e6,
 				minCacheableTokens: 1024,
 				requestPrice: 0,
 				discount: 0.2,
@@ -376,6 +383,8 @@ export const anthropicModels = [
 				inputPrice: 3.0 / 1e6,
 				outputPrice: 15.0 / 1e6,
 				cachedInputPrice: 0.3 / 1e6,
+				cacheWriteInputPrice: 3.75 / 1e6,
+				cacheWriteInputPrice1h: 6.0 / 1e6,
 				minCacheableTokens: 2048,
 				requestPrice: 0,
 				discount: 0.2,
@@ -443,6 +452,8 @@ export const anthropicModels = [
 				inputPrice: 1.0 / 1e6,
 				outputPrice: 5.0 / 1e6,
 				cachedInputPrice: 0.1 / 1e6,
+				cacheWriteInputPrice: 1.25 / 1e6,
+				cacheWriteInputPrice1h: 2.0 / 1e6,
 				minCacheableTokens: 4096,
 				requestPrice: 0,
 				discount: 0.2,
@@ -490,6 +501,8 @@ export const anthropicModels = [
 				inputPrice: 1.0 / 1e6,
 				outputPrice: 5.0 / 1e6,
 				cachedInputPrice: 0.1 / 1e6,
+				cacheWriteInputPrice: 1.25 / 1e6,
+				cacheWriteInputPrice1h: 2.0 / 1e6,
 				minCacheableTokens: 4096,
 				requestPrice: 0,
 				discount: 0.2,
@@ -540,6 +553,7 @@ export const anthropicModels = [
 				inputPrice: 15.0 / 1e6,
 				outputPrice: 75.0 / 1e6,
 				cachedInputPrice: 1.5 / 1e6,
+				cacheWriteInputPrice: 18.75 / 1e6,
 				minCacheableTokens: 1024,
 				requestPrice: 0,
 				discount: 0.2,
@@ -590,6 +604,7 @@ export const anthropicModels = [
 				inputPrice: 15.0 / 1e6,
 				outputPrice: 75.0 / 1e6,
 				cachedInputPrice: 1.5 / 1e6,
+				cacheWriteInputPrice: 18.75 / 1e6,
 				minCacheableTokens: 1024,
 				requestPrice: 0,
 				discount: 0.2,
@@ -698,6 +713,7 @@ export const anthropicModels = [
 				inputPrice: 0.8 / 1e6,
 				outputPrice: 4.0 / 1e6,
 				cachedInputPrice: 0.08 / 1e6,
+				cacheWriteInputPrice: 1.0 / 1e6,
 				minCacheableTokens: 2048,
 				requestPrice: 0,
 				discount: 0.2,
@@ -830,6 +846,8 @@ export const anthropicModels = [
 				inputPrice: 5.0 / 1e6,
 				outputPrice: 25.0 / 1e6,
 				cachedInputPrice: 0.5 / 1e6,
+				cacheWriteInputPrice: 6.25 / 1e6,
+				cacheWriteInputPrice1h: 10.0 / 1e6,
 				minCacheableTokens: 4096,
 				requestPrice: 0,
 				discount: 0.2,
@@ -921,6 +939,8 @@ export const anthropicModels = [
 				inputPrice: 5.0 / 1e6,
 				outputPrice: 25.0 / 1e6,
 				cachedInputPrice: 0.5 / 1e6,
+				cacheWriteInputPrice: 6.25 / 1e6,
+				cacheWriteInputPrice1h: 10.0 / 1e6,
 				minCacheableTokens: 4096,
 				requestPrice: 0,
 				discount: 0.2,
@@ -994,6 +1014,8 @@ export const anthropicModels = [
 				inputPrice: 5.0 / 1e6,
 				outputPrice: 25.0 / 1e6,
 				cachedInputPrice: 0.5 / 1e6,
+				cacheWriteInputPrice: 6.25 / 1e6,
+				cacheWriteInputPrice1h: 10.0 / 1e6,
 				minCacheableTokens: 4096,
 				requestPrice: 0,
 				discount: 0.2,