@ai-sdk/cohere 2.0.8 → 2.0.10

package/CHANGELOG.md

@@ -1,5 +1,20 @@
 # @ai-sdk/cohere
 
+## 2.0.10
+
+### Patch Changes
+
+- Updated dependencies [0294b58]
+  - @ai-sdk/provider-utils@3.0.9
+
+## 2.0.9
+
+### Patch Changes
+
+- 0816d3a: feat(provider/cohere): reasoning model support
+
+  Reasoning is now supported for all Cohere models that support it (`command-a-reasoning-08-2025` as of today). See https://docs.cohere.com/docs/reasoning
+
 ## 2.0.8
 
 ### Patch Changes

package/dist/index.d.mts

@@ -1,7 +1,18 @@
 import { ProviderV2, LanguageModelV2, EmbeddingModelV2 } from '@ai-sdk/provider';
 import { FetchFunction } from '@ai-sdk/provider-utils';
+import { z } from 'zod/v4';
 
-type CohereChatModelId = 'command-a-03-2025' | 'command-r7b-12-2024' | 'command-r-plus-04-2024' | 'command-r-plus' | 'command-r-08-2024' | 'command-r-03-2024' | 'command-r' | 'command' | 'command-nightly' | 'command-light' | 'command-light-nightly' | (string & {});
+type CohereChatModelId = 'command-a-03-2025' | 'command-a-reasoning-08-2025' | 'command-r7b-12-2024' | 'command-r-plus-04-2024' | 'command-r-plus' | 'command-r-08-2024' | 'command-r-03-2024' | 'command-r' | 'command' | 'command-nightly' | 'command-light' | 'command-light-nightly' | (string & {});
+declare const cohereChatModelOptions: z.ZodObject<{
+    thinking: z.ZodOptional<z.ZodObject<{
+        type: z.ZodOptional<z.ZodEnum<{
+            enabled: "enabled";
+            disabled: "disabled";
+        }>>;
+        tokenBudget: z.ZodOptional<z.ZodNumber>;
+    }, z.core.$strip>>;
+}, z.core.$strip>;
+type CohereChatModelOptions = z.infer<typeof cohereChatModelOptions>;
 
 type CohereEmbeddingModelId = 'embed-english-v3.0' | 'embed-multilingual-v3.0' | 'embed-english-light-v3.0' | 'embed-multilingual-light-v3.0' | 'embed-english-v2.0' | 'embed-english-light-v2.0' | 'embed-multilingual-v2.0' | (string & {});
 
@@ -48,4 +59,4 @@ Default Cohere provider instance.
  */
 declare const cohere: CohereProvider;
 
-export { type CohereProvider, type CohereProviderSettings, cohere, createCohere };
+export { type CohereChatModelOptions, type CohereProvider, type CohereProviderSettings, cohere, createCohere };

package/dist/index.d.ts

@@ -1,7 +1,18 @@
 import { ProviderV2, LanguageModelV2, EmbeddingModelV2 } from '@ai-sdk/provider';
 import { FetchFunction } from '@ai-sdk/provider-utils';
+import { z } from 'zod/v4';
 
-type CohereChatModelId = 'command-a-03-2025' | 'command-r7b-12-2024' | 'command-r-plus-04-2024' | 'command-r-plus' | 'command-r-08-2024' | 'command-r-03-2024' | 'command-r' | 'command' | 'command-nightly' | 'command-light' | 'command-light-nightly' | (string & {});
+type CohereChatModelId = 'command-a-03-2025' | 'command-a-reasoning-08-2025' | 'command-r7b-12-2024' | 'command-r-plus-04-2024' | 'command-r-plus' | 'command-r-08-2024' | 'command-r-03-2024' | 'command-r' | 'command' | 'command-nightly' | 'command-light' | 'command-light-nightly' | (string & {});
+declare const cohereChatModelOptions: z.ZodObject<{
+    thinking: z.ZodOptional<z.ZodObject<{
+        type: z.ZodOptional<z.ZodEnum<{
+            enabled: "enabled";
+            disabled: "disabled";
+        }>>;
+        tokenBudget: z.ZodOptional<z.ZodNumber>;
+    }, z.core.$strip>>;
+}, z.core.$strip>;
+type CohereChatModelOptions = z.infer<typeof cohereChatModelOptions>;
 
 type CohereEmbeddingModelId = 'embed-english-v3.0' | 'embed-multilingual-v3.0' | 'embed-english-light-v3.0' | 'embed-multilingual-light-v3.0' | 'embed-english-v2.0' | 'embed-english-light-v2.0' | 'embed-multilingual-v2.0' | (string & {});
 
@@ -48,4 +59,4 @@ Default Cohere provider instance.
  */
 declare const cohere: CohereProvider;
 
-export { type CohereProvider, type CohereProviderSettings, cohere, createCohere };
+export { type CohereChatModelOptions, type CohereProvider, type CohereProviderSettings, cohere, createCohere };

package/dist/index.js

@@ -31,13 +31,30 @@ var import_provider_utils4 = require("@ai-sdk/provider-utils");
 
 // src/cohere-chat-language-model.ts
 var import_provider_utils2 = require("@ai-sdk/provider-utils");
-var import_v42 = require("zod/v4");
+var import_v43 = require("zod/v4");
+
+// src/cohere-chat-options.ts
+var import_v4 = require("zod/v4");
+var cohereChatModelOptions = import_v4.z.object({
+  /**
+   * Configuration for reasoning features (optional)
+   *
+   * Can be set to an object with the two properties `type` and `tokenBudget`. `type` can be set to `'enabled'` or `'disabled'` (defaults to `'enabled'`).
+   * `tokenBudget` is the maximum number of tokens the model can use for thinking, which must be set to a positive integer. The model will stop thinking if it reaches the thinking token budget and will proceed with the response
+   *
+   * @see https://docs.cohere.com/reference/chat#request.body.thinking
+   */
+  thinking: import_v4.z.object({
+    type: import_v4.z.enum(["enabled", "disabled"]).optional(),
+    tokenBudget: import_v4.z.number().optional()
+  }).optional()
+});
 
 // src/cohere-error.ts
 var import_provider_utils = require("@ai-sdk/provider-utils");
-var import_v4 = require("zod/v4");
-var cohereErrorDataSchema = import_v4.z.object({
-  message: import_v4.z.string()
+var import_v42 = require("zod/v4");
+var cohereErrorDataSchema = import_v42.z.object({
+  message: import_v42.z.string()
 });
 var cohereFailedResponseHandler = (0, import_provider_utils.createJsonErrorResponseHandler)({
   errorSchema: cohereErrorDataSchema,
@@ -244,7 +261,7 @@ var CohereChatLanguageModel = class {
   get provider() {
     return this.config.provider;
   }
-  getArgs({
+  async getArgs({
     prompt,
     maxOutputTokens,
     temperature,
@@ -256,8 +273,15 @@ var CohereChatLanguageModel = class {
     responseFormat,
     seed,
     tools,
-    toolChoice
+    toolChoice,
+    providerOptions
   }) {
+    var _a, _b;
+    const cohereOptions = (_a = await (0, import_provider_utils2.parseProviderOptions)({
+      provider: "cohere",
+      providerOptions,
+      schema: cohereChatModelOptions
+    })) != null ? _a : {};
     const {
       messages: chatPrompt,
       documents: cohereDocuments,
@@ -289,14 +313,21 @@ var CohereChatLanguageModel = class {
         tools: cohereTools,
         tool_choice: cohereToolChoice,
         // documents for RAG:
-        ...cohereDocuments.length > 0 && { documents: cohereDocuments }
+        ...cohereDocuments.length > 0 && { documents: cohereDocuments },
+        // reasoning
+        ...cohereOptions.thinking && {
+          thinking: {
+            type: (_b = cohereOptions.thinking.type) != null ? _b : "enabled",
+            token_budget: cohereOptions.thinking.tokenBudget
+          }
+        }
       },
       warnings: [...toolWarnings, ...promptWarnings]
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c, _d, _e, _f, _g, _h, _i;
-    const { args, warnings } = this.getArgs(options);
+    var _a, _b, _c, _d, _e, _f;
+    const { args, warnings } = await this.getArgs(options);
     const {
       responseHeaders,
       value: response,
@@ -313,16 +344,23 @@ var CohereChatLanguageModel = class {
       fetch: this.config.fetch
     });
     const content = [];
-    if (((_b = (_a = response.message.content) == null ? void 0 : _a[0]) == null ? void 0 : _b.text) != null && ((_d = (_c = response.message.content) == null ? void 0 : _c[0]) == null ? void 0 : _d.text.length) > 0) {
-      content.push({ type: "text", text: response.message.content[0].text });
+    for (const item of (_a = response.message.content) != null ? _a : []) {
+      if (item.type === "text" && item.text.length > 0) {
+        content.push({ type: "text", text: item.text });
+        continue;
+      }
+      if (item.type === "thinking" && item.thinking.length > 0) {
+        content.push({ type: "reasoning", text: item.thinking });
+        continue;
+      }
     }
-    for (const citation of (_e = response.message.citations) != null ? _e : []) {
+    for (const citation of (_b = response.message.citations) != null ? _b : []) {
       content.push({
         type: "source",
         sourceType: "document",
         id: this.config.generateId(),
         mediaType: "text/plain",
-        title: ((_g = (_f = citation.sources[0]) == null ? void 0 : _f.document) == null ? void 0 : _g.title) || "Document",
+        title: ((_d = (_c = citation.sources[0]) == null ? void 0 : _c.document) == null ? void 0 : _d.title) || "Document",
         providerMetadata: {
           cohere: {
             start: citation.start,
@@ -334,7 +372,7 @@ var CohereChatLanguageModel = class {
         }
       });
     }
-    for (const toolCall of (_h = response.message.tool_calls) != null ? _h : []) {
+    for (const toolCall of (_e = response.message.tool_calls) != null ? _e : []) {
       content.push({
         type: "tool-call",
         toolCallId: toolCall.id,
@@ -355,7 +393,7 @@ var CohereChatLanguageModel = class {
       request: { body: args },
       response: {
         // TODO timestamp, model id
-        id: (_i = response.generation_id) != null ? _i : void 0,
+        id: (_f = response.generation_id) != null ? _f : void 0,
         headers: responseHeaders,
         body: rawResponse
       },
@@ -363,7 +401,7 @@ var CohereChatLanguageModel = class {
     };
   }
   async doStream(options) {
-    const { args, warnings } = this.getArgs(options);
+    const { args, warnings } = await this.getArgs(options);
     const { responseHeaders, value: response } = await (0, import_provider_utils2.postJsonToApi)({
       url: `${this.config.baseURL}/chat`,
       headers: (0, import_provider_utils2.combineHeaders)(this.config.headers(), options.headers),
@@ -382,6 +420,7 @@ var CohereChatLanguageModel = class {
       totalTokens: void 0
     };
     let pendingToolCall = null;
+    let isActiveReasoning = false;
     return {
       stream: response.pipeThrough(
         new TransformStream({
@@ -402,6 +441,14 @@ var CohereChatLanguageModel = class {
             const type = value.type;
             switch (type) {
               case "content-start": {
+                if (value.delta.message.content.type === "thinking") {
+                  controller.enqueue({
+                    type: "reasoning-start",
+                    id: String(value.index)
+                  });
+                  isActiveReasoning = true;
+                  return;
+                }
                 controller.enqueue({
                   type: "text-start",
                   id: String(value.index)
@@ -409,6 +456,14 @@ var CohereChatLanguageModel = class {
                 return;
               }
               case "content-delta": {
+                if ("thinking" in value.delta.message.content) {
+                  controller.enqueue({
+                    type: "reasoning-delta",
+                    id: String(value.index),
+                    delta: value.delta.message.content.thinking
+                  });
+                  return;
+                }
                 controller.enqueue({
                   type: "text-delta",
                   id: String(value.index),
@@ -417,6 +472,14 @@ var CohereChatLanguageModel = class {
                 return;
               }
               case "content-end": {
+                if (isActiveReasoning) {
+                  controller.enqueue({
+                    type: "reasoning-end",
+                    id: String(value.index)
+                  });
+                  isActiveReasoning = false;
+                  return;
+                }
                 controller.enqueue({
                   type: "text-end",
                   id: String(value.index)
@@ -512,120 +575,145 @@ var CohereChatLanguageModel = class {
     };
   }
 };
-var cohereChatResponseSchema = import_v42.z.object({
-  generation_id: import_v42.z.string().nullish(),
-  message: import_v42.z.object({
-    role: import_v42.z.string(),
-    content: import_v42.z.array(
-      import_v42.z.object({
-        type: import_v42.z.string(),
-        text: import_v42.z.string()
-      })
+var cohereChatResponseSchema = import_v43.z.object({
+  generation_id: import_v43.z.string().nullish(),
+  message: import_v43.z.object({
+    role: import_v43.z.string(),
+    content: import_v43.z.array(
+      import_v43.z.union([
+        import_v43.z.object({
+          type: import_v43.z.literal("text"),
+          text: import_v43.z.string()
+        }),
+        import_v43.z.object({
+          type: import_v43.z.literal("thinking"),
+          thinking: import_v43.z.string()
+        })
+      ])
     ).nullish(),
-    tool_plan: import_v42.z.string().nullish(),
-    tool_calls: import_v42.z.array(
-      import_v42.z.object({
-        id: import_v42.z.string(),
-        type: import_v42.z.literal("function"),
-        function: import_v42.z.object({
-          name: import_v42.z.string(),
-          arguments: import_v42.z.string()
+    tool_plan: import_v43.z.string().nullish(),
+    tool_calls: import_v43.z.array(
+      import_v43.z.object({
+        id: import_v43.z.string(),
+        type: import_v43.z.literal("function"),
+        function: import_v43.z.object({
+          name: import_v43.z.string(),
+          arguments: import_v43.z.string()
         })
       })
     ).nullish(),
-    citations: import_v42.z.array(
-      import_v42.z.object({
-        start: import_v42.z.number(),
-        end: import_v42.z.number(),
-        text: import_v42.z.string(),
-        sources: import_v42.z.array(
-          import_v42.z.object({
-            type: import_v42.z.string().optional(),
-            id: import_v42.z.string().optional(),
-            document: import_v42.z.object({
-              id: import_v42.z.string().optional(),
-              text: import_v42.z.string(),
-              title: import_v42.z.string()
+    citations: import_v43.z.array(
+      import_v43.z.object({
+        start: import_v43.z.number(),
+        end: import_v43.z.number(),
+        text: import_v43.z.string(),
+        sources: import_v43.z.array(
+          import_v43.z.object({
+            type: import_v43.z.string().optional(),
+            id: import_v43.z.string().optional(),
+            document: import_v43.z.object({
+              id: import_v43.z.string().optional(),
+              text: import_v43.z.string(),
+              title: import_v43.z.string()
             })
           })
         ),
-        type: import_v42.z.string().optional()
+        type: import_v43.z.string().optional()
       })
     ).nullish()
   }),
-  finish_reason: import_v42.z.string(),
-  usage: import_v42.z.object({
-    billed_units: import_v42.z.object({
-      input_tokens: import_v42.z.number(),
-      output_tokens: import_v42.z.number()
+  finish_reason: import_v43.z.string(),
+  usage: import_v43.z.object({
+    billed_units: import_v43.z.object({
+      input_tokens: import_v43.z.number(),
+      output_tokens: import_v43.z.number()
     }),
-    tokens: import_v42.z.object({
-      input_tokens: import_v42.z.number(),
-      output_tokens: import_v42.z.number()
+    tokens: import_v43.z.object({
+      input_tokens: import_v43.z.number(),
+      output_tokens: import_v43.z.number()
     })
   })
 });
-var cohereChatChunkSchema = import_v42.z.discriminatedUnion("type", [
-  import_v42.z.object({
-    type: import_v42.z.literal("citation-start")
+var cohereChatChunkSchema = import_v43.z.discriminatedUnion("type", [
+  import_v43.z.object({
+    type: import_v43.z.literal("citation-start")
   }),
-  import_v42.z.object({
-    type: import_v42.z.literal("citation-end")
+  import_v43.z.object({
+    type: import_v43.z.literal("citation-end")
   }),
-  import_v42.z.object({
-    type: import_v42.z.literal("content-start"),
-    index: import_v42.z.number()
+  import_v43.z.object({
+    type: import_v43.z.literal("content-start"),
+    index: import_v43.z.number(),
+    delta: import_v43.z.object({
+      message: import_v43.z.object({
+        content: import_v43.z.union([
+          import_v43.z.object({
+            type: import_v43.z.literal("text"),
+            text: import_v43.z.string()
+          }),
+          import_v43.z.object({
+            type: import_v43.z.literal("thinking"),
+            thinking: import_v43.z.string()
+          })
+        ])
+      })
+    })
   }),
-  import_v42.z.object({
-    type: import_v42.z.literal("content-delta"),
-    index: import_v42.z.number(),
-    delta: import_v42.z.object({
-      message: import_v42.z.object({
-        content: import_v42.z.object({
-          text: import_v42.z.string()
-        })
+  import_v43.z.object({
+    type: import_v43.z.literal("content-delta"),
+    index: import_v43.z.number(),
+    delta: import_v43.z.object({
+      message: import_v43.z.object({
+        content: import_v43.z.union([
+          import_v43.z.object({
+            text: import_v43.z.string()
+          }),
+          import_v43.z.object({
+            thinking: import_v43.z.string()
+          })
+        ])
       })
     })
   }),
-  import_v42.z.object({
-    type: import_v42.z.literal("content-end"),
-    index: import_v42.z.number()
+  import_v43.z.object({
+    type: import_v43.z.literal("content-end"),
+    index: import_v43.z.number()
   }),
-  import_v42.z.object({
-    type: import_v42.z.literal("message-start"),
-    id: import_v42.z.string().nullish()
+  import_v43.z.object({
+    type: import_v43.z.literal("message-start"),
+    id: import_v43.z.string().nullish()
   }),
-  import_v42.z.object({
-    type: import_v42.z.literal("message-end"),
-    delta: import_v42.z.object({
-      finish_reason: import_v42.z.string(),
-      usage: import_v42.z.object({
-        tokens: import_v42.z.object({
-          input_tokens: import_v42.z.number(),
-          output_tokens: import_v42.z.number()
+  import_v43.z.object({
+    type: import_v43.z.literal("message-end"),
+    delta: import_v43.z.object({
+      finish_reason: import_v43.z.string(),
+      usage: import_v43.z.object({
+        tokens: import_v43.z.object({
+          input_tokens: import_v43.z.number(),
+          output_tokens: import_v43.z.number()
         })
       })
     })
   }),
   // https://docs.cohere.com/v2/docs/streaming#tool-use-stream-events-for-tool-calling
-  import_v42.z.object({
-    type: import_v42.z.literal("tool-plan-delta"),
-    delta: import_v42.z.object({
-      message: import_v42.z.object({
-        tool_plan: import_v42.z.string()
+  import_v43.z.object({
+    type: import_v43.z.literal("tool-plan-delta"),
+    delta: import_v43.z.object({
+      message: import_v43.z.object({
+        tool_plan: import_v43.z.string()
       })
     })
   }),
-  import_v42.z.object({
-    type: import_v42.z.literal("tool-call-start"),
-    delta: import_v42.z.object({
-      message: import_v42.z.object({
-        tool_calls: import_v42.z.object({
-          id: import_v42.z.string(),
-          type: import_v42.z.literal("function"),
-          function: import_v42.z.object({
-            name: import_v42.z.string(),
-            arguments: import_v42.z.string()
+  import_v43.z.object({
+    type: import_v43.z.literal("tool-call-start"),
+    delta: import_v43.z.object({
+      message: import_v43.z.object({
+        tool_calls: import_v43.z.object({
+          id: import_v43.z.string(),
+          type: import_v43.z.literal("function"),
+          function: import_v43.z.object({
+            name: import_v43.z.string(),
+            arguments: import_v43.z.string()
           })
         })
       })
@@ -634,31 +722,31 @@ var cohereChatChunkSchema = import_v42.z.discriminatedUnion("type", [
   // A single tool call's `arguments` stream in chunks and must be accumulated
   // in a string and so the full tool object info can only be parsed once we see
   // `tool-call-end`.
-  import_v42.z.object({
-    type: import_v42.z.literal("tool-call-delta"),
-    delta: import_v42.z.object({
-      message: import_v42.z.object({
-        tool_calls: import_v42.z.object({
-          function: import_v42.z.object({
-            arguments: import_v42.z.string()
+  import_v43.z.object({
+    type: import_v43.z.literal("tool-call-delta"),
+    delta: import_v43.z.object({
+      message: import_v43.z.object({
+        tool_calls: import_v43.z.object({
+          function: import_v43.z.object({
+            arguments: import_v43.z.string()
           })
         })
       })
     })
   }),
-  import_v42.z.object({
-    type: import_v42.z.literal("tool-call-end")
+  import_v43.z.object({
+    type: import_v43.z.literal("tool-call-end")
   })
 ]);
 
 // src/cohere-embedding-model.ts
 var import_provider3 = require("@ai-sdk/provider");
 var import_provider_utils3 = require("@ai-sdk/provider-utils");
-var import_v44 = require("zod/v4");
+var import_v45 = require("zod/v4");
 
 // src/cohere-embedding-options.ts
-var import_v43 = require("zod/v4");
-var cohereEmbeddingOptions = import_v43.z.object({
+var import_v44 = require("zod/v4");
+var cohereEmbeddingOptions = import_v44.z.object({
   /**
    * Specifies the type of input passed to the model. Default is `search_query`.
    *
@@ -667,7 +755,7 @@ var cohereEmbeddingOptions = import_v43.z.object({
    * - "classification": Used for embeddings passed through a text classifier.
    * - "clustering": Used for embeddings run through a clustering algorithm.
    */
-  inputType: import_v43.z.enum(["search_document", "search_query", "classification", "clustering"]).optional(),
+  inputType: import_v44.z.enum(["search_document", "search_query", "classification", "clustering"]).optional(),
   /**
    * Specifies how the API will handle inputs longer than the maximum token length.
    * Default is `END`.
@@ -676,7 +764,7 @@ var cohereEmbeddingOptions = import_v43.z.object({
    * - "START": Will discard the start of the input until the remaining input is exactly the maximum input token length for the model.
    * - "END": Will discard the end of the input until the remaining input is exactly the maximum input token length for the model.
    */
-  truncate: import_v43.z.enum(["NONE", "START", "END"]).optional()
+  truncate: import_v44.z.enum(["NONE", "START", "END"]).optional()
 });
 
 // src/cohere-embedding-model.ts
@@ -742,13 +830,13 @@ var CohereEmbeddingModel = class {
     };
   }
 };
-var cohereTextEmbeddingResponseSchema = import_v44.z.object({
-  embeddings: import_v44.z.object({
-    float: import_v44.z.array(import_v44.z.array(import_v44.z.number()))
+var cohereTextEmbeddingResponseSchema = import_v45.z.object({
+  embeddings: import_v45.z.object({
+    float: import_v45.z.array(import_v45.z.array(import_v45.z.number()))
   }),
-  meta: import_v44.z.object({
-    billed_units: import_v44.z.object({
-      input_tokens: import_v44.z.number()
+  meta: import_v45.z.object({
+    billed_units: import_v45.z.object({
+      input_tokens: import_v45.z.number()
     })
   })
 });