@ai-sdk/anthropic 2.0.57 → 2.0.59

package/dist/internal/index.mjs

@@ -497,8 +497,10 @@ var anthropicMessagesChunkSchema = lazySchema2(
           }).nullish()
         }),
         usage: z2.looseObject({
+          input_tokens: z2.number().nullish(),
           output_tokens: z2.number(),
-          cache_creation_input_tokens: z2.number().nullish()
+          cache_creation_input_tokens: z2.number().nullish(),
+          cache_read_input_tokens: z2.number().nullish()
         })
       }),
       z2.object({
@@ -560,10 +562,20 @@ var anthropicProviderOptions = z3.object({
    * When enabled, responses include thinking content blocks showing Claude's thinking process before the final answer.
    * Requires a minimum budget of 1,024 tokens and counts towards the `max_tokens` limit.
    */
-  thinking: z3.object({
-    type: z3.union([z3.literal("enabled"), z3.literal("disabled")]),
-    budgetTokens: z3.number().optional()
-  }).optional(),
+  thinking: z3.discriminatedUnion("type", [
+    z3.object({
+      /** for Opus 4.6 and newer models */
+      type: z3.literal("adaptive")
+    }),
+    z3.object({
+      /** for models before Opus 4.6 */
+      type: z3.literal("enabled"),
+      budgetTokens: z3.number().optional()
+    }),
+    z3.object({
+      type: z3.literal("disabled")
+    })
+  ]).optional(),
   /**
    * Whether to disable parallel function calling during tool use. Default is false.
    * When set to true, Claude will use at most one tool per response.
@@ -595,7 +607,7 @@ var anthropicProviderOptions = z3.object({
   /**
    * @default 'high'
    */
-  effort: z3.enum(["low", "medium", "high"]).optional()
+  effort: z3.enum(["low", "medium", "high", "max"]).optional()
 });
 
 // src/anthropic-prepare-tools.ts
@@ -1877,8 +1889,9 @@ var AnthropicMessagesLanguageModel = class {
       warnings,
       cacheControlValidator
     });
-    const isThinking = ((_c = anthropicOptions == null ? void 0 : anthropicOptions.thinking) == null ? void 0 : _c.type) === "enabled";
-    const thinkingBudget = (_d = anthropicOptions == null ? void 0 : anthropicOptions.thinking) == null ? void 0 : _d.budgetTokens;
+    const thinkingType = (_c = anthropicOptions == null ? void 0 : anthropicOptions.thinking) == null ? void 0 : _c.type;
+    const isThinking = thinkingType === "enabled" || thinkingType === "adaptive";
+    let thinkingBudget = thinkingType === "enabled" ? (_d = anthropicOptions == null ? void 0 : anthropicOptions.thinking) == null ? void 0 : _d.budgetTokens : void 0;
     const maxTokens = maxOutputTokens != null ? maxOutputTokens : maxOutputTokensForModel;
     const baseArgs = {
       // model id:
@@ -1891,7 +1904,10 @@ var AnthropicMessagesLanguageModel = class {
       stop_sequences: stopSequences,
       // provider specific settings:
       ...isThinking && {
-        thinking: { type: "enabled", budget_tokens: thinkingBudget }
+        thinking: {
+          type: thinkingType,
+          ...thinkingBudget != null && { budget_tokens: thinkingBudget }
+        }
       },
       ...(anthropicOptions == null ? void 0 : anthropicOptions.effort) && {
         output_config: { effort: anthropicOptions.effort }
@@ -1919,7 +1935,7 @@ var AnthropicMessagesLanguageModel = class {
       messages: messagesPrompt.messages
     };
     if (isThinking) {
-      if (thinkingBudget == null) {
+      if (thinkingType === "enabled" && thinkingBudget == null) {
         throw new UnsupportedFunctionalityError3({
           functionality: "thinking requires a budget"
         });
@@ -1948,7 +1964,7 @@ var AnthropicMessagesLanguageModel = class {
           details: "topP is not supported when thinking is enabled"
         });
       }
-      baseArgs.max_tokens = maxTokens + thinkingBudget;
+      baseArgs.max_tokens = maxTokens + (thinkingBudget != null ? thinkingBudget : 0);
     }
     if (isKnownModel && baseArgs.max_tokens > maxOutputTokensForModel) {
       if (maxOutputTokens != null) {
@@ -2745,8 +2761,17 @@ var AnthropicMessagesLanguageModel = class {
               return;
             }
             case "message_delta": {
+              if (value.usage.input_tokens != null && usage.inputTokens !== value.usage.input_tokens) {
+                usage.inputTokens = value.usage.input_tokens;
+              }
               usage.outputTokens = value.usage.output_tokens;
-              usage.totalTokens = ((_f = usage.inputTokens) != null ? _f : 0) + ((_g = value.usage.output_tokens) != null ? _g : 0);
+              if (value.usage.cache_read_input_tokens != null) {
+                usage.cachedInputTokens = value.usage.cache_read_input_tokens;
+              }
+              if (value.usage.cache_creation_input_tokens != null) {
+                cacheCreationInputTokens = value.usage.cache_creation_input_tokens;
+              }
+              usage.totalTokens = ((_f = usage.inputTokens) != null ? _f : 0) + ((_g = usage.outputTokens) != null ? _g : 0);
               finishReason = mapAnthropicStopReason({
                 finishReason: value.delta.stop_reason,
                 isJsonResponseFromTool: usesJsonResponseTool
@@ -2828,7 +2853,13 @@ var AnthropicMessagesLanguageModel = class {
   }
 };
 function getModelCapabilities(modelId) {
-  if (modelId.includes("claude-sonnet-4-5") || modelId.includes("claude-opus-4-5")) {
+  if (modelId.includes("claude-opus-4-6")) {
+    return {
+      maxOutputTokens: 128e3,
+      supportsStructuredOutput: true,
+      isKnownModel: true
+    };
+  } else if (modelId.includes("claude-sonnet-4-5") || modelId.includes("claude-opus-4-5")) {
     return {
       maxOutputTokens: 64e3,
       supportsStructuredOutput: true,