diff --git a/dist/index.js b/dist/index.js
index ce8d4ae802f5909937840b856866c2d735800212..d448e921d51cb0a4f752cdad09080b21d4d8e8a9 100644
--- a/dist/index.js
+++ b/dist/index.js
@@ -139,11 +139,12 @@ function convertCohereUsage(tokens) {
   }
   const inputTokens = tokens.input_tokens;
   const outputTokens = tokens.output_tokens;
+  const cacheReadTokens = tokens.cached_tokens ?? 0;
   return {
     inputTokens: {
       total: inputTokens,
-      noCache: inputTokens,
-      cacheRead: void 0,
+      noCache: Math.max(0, inputTokens - cacheReadTokens),
+      cacheRead: cacheReadTokens || void 0,
       cacheWrite: void 0
     },
     outputTokens: {
@@ -672,7 +673,8 @@ var cohereChatResponseSchema = import_v43.z.object({
     }),
     tokens: import_v43.z.object({
       input_tokens: import_v43.z.number(),
-      output_tokens: import_v43.z.number()
+      output_tokens: import_v43.z.number(),
+      cached_tokens: import_v43.z.number().optional()
     })
   })
 });
@@ -732,7 +734,8 @@ var cohereChatChunkSchema = import_v43.z.discriminatedUnion("type", [
       usage: import_v43.z.object({
         tokens: import_v43.z.object({
           input_tokens: import_v43.z.number(),
-          output_tokens: import_v43.z.number()
+          output_tokens: import_v43.z.number(),
+          cached_tokens: import_v43.z.number().optional()
         })
       })
     })
diff --git a/dist/index.mjs b/dist/index.mjs
index 44f26c27511f9089cdcc9456590dee100b070b0a..d3f585b58208430404135060af5774a7ecfa65d0 100644
--- a/dist/index.mjs
+++ b/dist/index.mjs
@@ -126,11 +126,12 @@ function convertCohereUsage(tokens) {
   }
   const inputTokens = tokens.input_tokens;
   const outputTokens = tokens.output_tokens;
+  const cacheReadTokens = tokens.cached_tokens ?? 0;
   return {
     inputTokens: {
       total: inputTokens,
-      noCache: inputTokens,
-      cacheRead: void 0,
+      noCache: Math.max(0, inputTokens - cacheReadTokens),
+      cacheRead: cacheReadTokens || void 0,
       cacheWrite: void 0
     },
     outputTokens: {
@@ -661,7 +662,8 @@ var cohereChatResponseSchema = z3.object({
     }),
     tokens: z3.object({
       input_tokens: z3.number(),
-      output_tokens: z3.number()
+      output_tokens: z3.number(),
+      cached_tokens: z3.number().optional()
     })
   })
 });
@@ -721,7 +723,8 @@ var cohereChatChunkSchema = z3.discriminatedUnion("type", [
       usage: z3.object({
         tokens: z3.object({
           input_tokens: z3.number(),
-          output_tokens: z3.number()
+          output_tokens: z3.number(),
+          cached_tokens: z3.number().optional()
         })
       })
     })
diff --git a/src/cohere-chat-language-model.ts b/src/cohere-chat-language-model.ts
index e4971ccf255ffff77c38fd35caccf84a032ad150..93fea74a08c0f9bcc6c63689fbc9e72669cdeba1 100644
--- a/src/cohere-chat-language-model.ts
+++ b/src/cohere-chat-language-model.ts
@@ -495,6 +495,7 @@ const cohereChatResponseSchema = z.object({
     tokens: z.object({
       input_tokens: z.number(),
       output_tokens: z.number(),
+      cached_tokens: z.number().optional(),
     }),
   }),
 });
@@ -558,6 +559,7 @@ const cohereChatChunkSchema = z.discriminatedUnion('type', [
         tokens: z.object({
           input_tokens: z.number(),
           output_tokens: z.number(),
+          cached_tokens: z.number().optional(),
         }),
       }),
     }),
diff --git a/src/convert-cohere-usage.ts b/src/convert-cohere-usage.ts
index 0f83a63ab85ab2c2ae136edd509e1f738553041f..3913f8d221c8980916769a9890b56030f69a6070 100644
--- a/src/convert-cohere-usage.ts
+++ b/src/convert-cohere-usage.ts
@@ -3,6 +3,7 @@ import { LanguageModelV3Usage } from '@ai-sdk/provider';
 export type CohereUsageTokens = {
   input_tokens: number;
   output_tokens: number;
+  cached_tokens?: number;
 };
 
 export function convertCohereUsage(
@@ -27,12 +28,13 @@ export function convertCohereUsage(
 
   const inputTokens = tokens.input_tokens;
   const outputTokens = tokens.output_tokens;
+  const cacheReadTokens = tokens.cached_tokens ?? 0;
 
   return {
     inputTokens: {
       total: inputTokens,
-      noCache: inputTokens,
-      cacheRead: undefined,
+      noCache: Math.max(0, inputTokens - cacheReadTokens),
+      cacheRead: cacheReadTokens || undefined,
       cacheWrite: undefined,
     },
     outputTokens: {