@ai-sdk/anthropic 2.0.89 → 2.0.91

package/dist/index.mjs

@@ -11,7 +11,7 @@ import {
 } from "@ai-sdk/provider-utils";
 
 // src/version.ts
-var VERSION = true ? "2.0.89" : "0.0.0-test";
+var VERSION = true ? "2.0.91" : "0.0.0-test";
 
 // src/anthropic-messages-language-model.ts
 import {
@@ -662,7 +662,40 @@ var anthropicFilePartProviderOptions = z3.object({
    */
   context: z3.string().optional()
 });
+var anthropicSystemMessageProviderOptions = z3.object({
+  /**
+   * Mid-conversation tool changes. Adds or removes tools from the
+   * conversation's tool set between turns without invalidating the prompt
+   * cache.
+   *
+   * Only supported on system messages that appear mid-conversation (not the
+   * initial system prompt). A system message carrying tool changes must come
+   * right before an assistant message or at the end of the messages.
+   *
+   * Tools referenced by a `tool_addition` must be declared in the `tools`
+   * option (typically with `deferLoading: true` so they are not loaded until
+   * the addition surfaces them). The required
+   * `mid-conversation-tool-changes-2026-07-01` beta is added automatically.
+   */
+  toolChanges: z3.array(
+    z3.discriminatedUnion("type", [
+      z3.object({
+        type: z3.literal("tool_addition"),
+        toolName: z3.string()
+      }),
+      z3.object({
+        type: z3.literal("tool_removal"),
+        toolName: z3.string()
+      })
+    ])
+  ).optional()
+});
 var anthropicProviderOptions = z3.object({
+  /**
+   * Whether to send reasoning to the model.
+   *
+   * This allows you to deactivate reasoning inputs for models that do not support them.
+   */
   sendReasoning: z3.boolean().optional(),
   /**
    * Determines how structured outputs are generated.
@@ -771,30 +804,35 @@ var anthropicProviderOptions = z3.object({
    */
   inferenceGeo: z3.enum(["us", "global"]).optional(),
   /**
-   * Server-side fallback chain.
+   * Server-side fallback configuration.
    *
    * When the primary model's safety classifiers block a turn, the API
-   * automatically retries it on the next model in the chain, server-side. A
-   * `content-filter` finish reason means the entire chain refused.
-   *
-   * Each entry is merged into the request as a direct request to that entry's
-   * model, so it must be formatted accordingly: `model` is required, and an
-   * entry may additionally override `max_tokens`, `thinking`, `output_config`,
-   * and `speed` for that attempt only (`speed` additionally requires the speed
-   * beta). The value is passed through to the API as-is.
+   * automatically retries it server-side on a fallback model. A
+   * `content-filter` finish reason means the fallback(s) refused as well.
    *
-   * The required `server-side-fallback-2026-06-01` beta is added automatically
-   * when this option is set.
+   * - `'default'` (recommended): the API routes the retry to Anthropic's
+   *   recommended fallback model based on the refusal category. Requires the
+   *   `server-side-fallback-2026-07-01` beta, which is added automatically.
+   * - Array form: an explicit fallback chain. Each entry is merged into the
+   *   request as a direct request to that entry's model, so it must be
+   *   formatted accordingly: `model` is required, and an entry may
+   *   additionally override `max_tokens`, `thinking`, `output_config`, and
+   *   `speed` for that attempt only (`speed` additionally requires the speed
+   *   beta). The value is passed through to the API as-is, and the
+   *   `server-side-fallback-2026-06-01` beta is added automatically.
    */
-  fallbacks: z3.array(
-    z3.object({
-      model: z3.string(),
-      max_tokens: z3.number().int().optional(),
-      thinking: z3.record(z3.string(), z3.unknown()).optional(),
-      output_config: z3.record(z3.string(), z3.unknown()).optional(),
-      speed: z3.enum(["fast", "standard"]).optional()
-    })
-  ).optional(),
+  fallbacks: z3.union([
+    z3.literal("default"),
+    z3.array(
+      z3.object({
+        model: z3.string(),
+        max_tokens: z3.number().int().optional(),
+        thinking: z3.record(z3.string(), z3.unknown()).optional(),
+        output_config: z3.record(z3.string(), z3.unknown()).optional(),
+        speed: z3.enum(["fast", "standard"]).optional()
+      })
+    )
+  ]).optional(),
   /**
    * Context management configuration for automatic context window management.
    * Enables features like automatic compaction and clearing of tool uses/thinking blocks.
@@ -1461,7 +1499,7 @@ async function convertToAnthropicMessagesPrompt({
   warnings,
   cacheControlValidator
 }) {
-  var _a, _b, _c, _d, _e, _f;
+  var _a, _b, _c, _d, _e, _f, _g;
   const betas = /* @__PURE__ */ new Set();
   const blocks = groupIntoBlocks(prompt);
   const validator = cacheControlValidator || new CacheControlValidator();
@@ -1493,19 +1531,52 @@ async function convertToAnthropicMessagesPrompt({
     const type = block.type;
     switch (type) {
       case "system": {
-        const content = block.messages.map(({ content: content2, providerOptions }) => ({
-          type: "text",
-          text: content2,
-          cache_control: validator.getCacheControl(providerOptions, {
-            type: "system message",
-            canCache: true
-          })
-        }));
-        if (system == null) {
-          system = content;
+        const content = [];
+        let toolChangeCount = 0;
+        for (const { content: text, providerOptions } of block.messages) {
+          const systemMessageOptions = await parseProviderOptions({
+            provider: "anthropic",
+            providerOptions,
+            schema: anthropicSystemMessageProviderOptions
+          });
+          const toolChanges = (_a = systemMessageOptions == null ? void 0 : systemMessageOptions.toolChanges) != null ? _a : [];
+          if (text !== "" || toolChanges.length === 0) {
+            content.push({
+              type: "text",
+              text,
+              cache_control: validator.getCacheControl(providerOptions, {
+                type: "system message",
+                canCache: true
+              })
+            });
+          }
+          for (const toolChange of toolChanges) {
+            toolChangeCount++;
+            content.push({
+              type: toolChange.type,
+              tool: {
+                type: "tool_reference",
+                name: toolChange.toolName
+              }
+            });
+          }
+        }
+        if (i === 0 || system == null && toolChangeCount === 0) {
+          if (toolChangeCount > 0) {
+            warnings.push({
+              type: "other",
+              message: "tool changes on the initial system message are not supported by Anthropic. Configure the initial tool set via the tools option instead. The tool changes have been ignored."
+            });
+          }
+          system = content.filter(
+            (part) => part.type === "text"
+          );
         } else {
           messages.push({ role: "system", content });
           betas.add("mid-conversation-system-2026-04-07");
+          if (toolChangeCount > 0) {
+            betas.add("mid-conversation-tool-changes-2026-07-01");
+          }
         }
         break;
       }
@@ -1518,10 +1589,10 @@ async function convertToAnthropicMessagesPrompt({
               for (let j = 0; j < content.length; j++) {
                 const part = content[j];
                 const isLastPart = j === content.length - 1;
-                const cacheControl = (_a = validator.getCacheControl(part.providerOptions, {
+                const cacheControl = (_b = validator.getCacheControl(part.providerOptions, {
                   type: "user message part",
                   canCache: true
-                })) != null ? _a : isLastPart ? validator.getCacheControl(message.providerOptions, {
+                })) != null ? _b : isLastPart ? validator.getCacheControl(message.providerOptions, {
                   type: "user message",
                   canCache: true
                 }) : void 0;
@@ -1566,7 +1637,7 @@ async function convertToAnthropicMessagesPrompt({
                           media_type: "application/pdf",
                           data: convertToBase64(part.data)
                         },
-                        title: (_b = metadata.title) != null ? _b : part.filename,
+                        title: (_c = metadata.title) != null ? _c : part.filename,
                         ...metadata.context && { context: metadata.context },
                         ...enableCitations && {
                           citations: { enabled: true }
@@ -1590,7 +1661,7 @@ async function convertToAnthropicMessagesPrompt({
                           media_type: "text/plain",
                           data: convertToString(part.data)
                         },
-                        title: (_c = metadata.title) != null ? _c : part.filename,
+                        title: (_d = metadata.title) != null ? _d : part.filename,
                         ...metadata.context && { context: metadata.context },
                         ...enableCitations && {
                           citations: { enabled: true }
@@ -1612,10 +1683,10 @@ async function convertToAnthropicMessagesPrompt({
               for (let i2 = 0; i2 < content.length; i2++) {
                 const part = content[i2];
                 const isLastPart = i2 === content.length - 1;
-                const cacheControl = (_d = validator.getCacheControl(part.providerOptions, {
+                const cacheControl = (_e = validator.getCacheControl(part.providerOptions, {
                   type: "tool result part",
                   canCache: true
-                })) != null ? _d : isLastPart ? validator.getCacheControl(message.providerOptions, {
+                })) != null ? _e : isLastPart ? validator.getCacheControl(message.providerOptions, {
                   type: "tool result message",
                   canCache: true
                 }) : void 0;
@@ -1697,16 +1768,16 @@ async function convertToAnthropicMessagesPrompt({
           for (let k = 0; k < content.length; k++) {
             const part = content[k];
             const isLastContentPart = k === content.length - 1;
-            const cacheControl = (_e = validator.getCacheControl(part.providerOptions, {
+            const cacheControl = (_f = validator.getCacheControl(part.providerOptions, {
               type: "assistant message part",
               canCache: true
-            })) != null ? _e : isLastContentPart ? validator.getCacheControl(message.providerOptions, {
+            })) != null ? _f : isLastContentPart ? validator.getCacheControl(message.providerOptions, {
               type: "assistant message",
               canCache: true
             }) : void 0;
             switch (part.type) {
               case "text": {
-                const textMetadata = (_f = part.providerOptions) == null ? void 0 : _f.anthropic;
+                const textMetadata = (_g = part.providerOptions) == null ? void 0 : _g.anthropic;
                 if ((textMetadata == null ? void 0 : textMetadata.type) === "compaction") {
                   anthropicContent.push({
                     type: "compaction",
@@ -2233,7 +2304,7 @@ var AnthropicMessagesLanguageModel = class {
     toolChoice,
     providerOptions
   }) {
-    var _a, _b, _c, _d, _e, _f, _g, _h;
+    var _a, _b, _c, _d, _e, _f, _g, _h, _i;
     const warnings = [];
     if (frequencyPenalty != null) {
       warnings.push({
@@ -2292,6 +2363,7 @@ var AnthropicMessagesLanguageModel = class {
       maxOutputTokens: maxOutputTokensForModel,
       supportsStructuredOutput,
       rejectsSamplingParameters,
+      rejectsThinkingDisabledAboveHighEffort,
       isKnownModel
     } = getModelCapabilities(this.modelId);
     if (!isKnownModel && maxOutputTokens == null) {
@@ -2326,7 +2398,7 @@ var AnthropicMessagesLanguageModel = class {
         topP = void 0;
       }
     }
-    const isAnthropicModel = isKnownModel || this.modelId.startsWith("claude-");
+    const isAnthropicModel = isKnownModel || this.modelId.includes("claude-");
     const structureOutputMode = (_a = anthropicOptions == null ? void 0 : anthropicOptions.structuredOutputMode) != null ? _a : "jsonTool";
     const useStructuredOutput = structureOutputMode === "outputFormat" || structureOutputMode === "auto" && supportsStructuredOutput;
     const jsonResponseTool = (responseFormat == null ? void 0 : responseFormat.type) === "json" && responseFormat.schema != null && !useStructuredOutput ? {
@@ -2349,10 +2421,18 @@ var AnthropicMessagesLanguageModel = class {
       warnings,
       cacheControlValidator
     });
-    const thinkingType = (_c = anthropicOptions == null ? void 0 : anthropicOptions.thinking) == null ? void 0 : _c.type;
+    if (rejectsThinkingDisabledAboveHighEffort && ((_c = anthropicOptions == null ? void 0 : anthropicOptions.thinking) == null ? void 0 : _c.type) === "disabled" && (anthropicOptions.effort === "xhigh" || anthropicOptions.effort === "max")) {
+      warnings.push({
+        type: "unsupported-setting",
+        setting: "providerOptions.anthropic.effort",
+        details: `effort '${anthropicOptions.effort}' is not supported by ${this.modelId} when thinking is disabled. The effort has been lowered to 'high'.`
+      });
+      anthropicOptions.effort = "high";
+    }
+    const thinkingType = (_d = anthropicOptions == null ? void 0 : anthropicOptions.thinking) == null ? void 0 : _d.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 thinkingDisplay = thinkingType === "adaptive" ? (_e = anthropicOptions == null ? void 0 : anthropicOptions.thinking) == null ? void 0 : _e.display : void 0;
+    let thinkingBudget = thinkingType === "enabled" ? (_e = anthropicOptions == null ? void 0 : anthropicOptions.thinking) == null ? void 0 : _e.budgetTokens : void 0;
+    const thinkingDisplay = thinkingType === "adaptive" ? (_f = anthropicOptions == null ? void 0 : anthropicOptions.thinking) == null ? void 0 : _f.display : void 0;
     const maxTokens = maxOutputTokens != null ? maxOutputTokens : maxOutputTokensForModel;
     const baseArgs = {
       // model id:
@@ -2399,20 +2479,20 @@ var AnthropicMessagesLanguageModel = class {
       ...(anthropicOptions == null ? void 0 : anthropicOptions.inferenceGeo) && {
         inference_geo: anthropicOptions.inferenceGeo
       },
-      ...(anthropicOptions == null ? void 0 : anthropicOptions.fallbacks) && anthropicOptions.fallbacks.length > 0 && {
+      ...(anthropicOptions == null ? void 0 : anthropicOptions.fallbacks) != null && (anthropicOptions.fallbacks === "default" || anthropicOptions.fallbacks.length > 0) && {
         fallbacks: anthropicOptions.fallbacks
       },
       ...(anthropicOptions == null ? void 0 : anthropicOptions.cacheControl) && {
         cache_control: anthropicOptions.cacheControl
       },
-      ...((_f = anthropicOptions == null ? void 0 : anthropicOptions.metadata) == null ? void 0 : _f.userId) != null && {
+      ...((_g = anthropicOptions == null ? void 0 : anthropicOptions.metadata) == null ? void 0 : _g.userId) != null && {
         metadata: { user_id: anthropicOptions.metadata.userId }
       },
       // container with agent skills:
       ...(anthropicOptions == null ? void 0 : anthropicOptions.container) && {
         container: {
           id: anthropicOptions.container.id,
-          skills: (_g = anthropicOptions.container.skills) == null ? void 0 : _g.map((skill) => ({
+          skills: (_h = anthropicOptions.container.skills) == null ? void 0 : _h.map((skill) => ({
             type: skill.type,
             skill_id: skill.skillId,
             version: skill.version
@@ -2544,7 +2624,9 @@ var AnthropicMessagesLanguageModel = class {
     if ((anthropicOptions == null ? void 0 : anthropicOptions.speed) === "fast") {
       betas.add("fast-mode-2026-02-01");
     }
-    if ((anthropicOptions == null ? void 0 : anthropicOptions.fallbacks) && anthropicOptions.fallbacks.length > 0) {
+    if ((anthropicOptions == null ? void 0 : anthropicOptions.fallbacks) === "default") {
+      betas.add("server-side-fallback-2026-07-01");
+    } else if ((anthropicOptions == null ? void 0 : anthropicOptions.fallbacks) && anthropicOptions.fallbacks.length > 0) {
       betas.add("server-side-fallback-2026-06-01");
     }
     if (useStructuredOutput) {
@@ -2587,7 +2669,7 @@ var AnthropicMessagesLanguageModel = class {
         ...betas,
         ...toolsBetas,
         ...userSuppliedBetas,
-        ...(_h = anthropicOptions == null ? void 0 : anthropicOptions.anthropicBeta) != null ? _h : []
+        ...(_i = anthropicOptions == null ? void 0 : anthropicOptions.anthropicBeta) != null ? _i : []
       ]),
       usesJsonResponseTool: jsonResponseTool != null
     };
@@ -3603,11 +3685,20 @@ var AnthropicMessagesLanguageModel = class {
   }
 };
 function getModelCapabilities(modelId) {
-  if (modelId.includes("claude-opus-4-8") || modelId.includes("claude-opus-4-7") || modelId.includes("claude-fable-5") || modelId.includes("claude-sonnet-5")) {
+  if (modelId.includes("claude-opus-5")) {
     return {
       maxOutputTokens: 128e3,
       supportsStructuredOutput: true,
       rejectsSamplingParameters: true,
+      rejectsThinkingDisabledAboveHighEffort: true,
+      isKnownModel: true
+    };
+  } else if (modelId.includes("claude-opus-4-8") || modelId.includes("claude-opus-4-7") || modelId.includes("claude-fable-5") || modelId.includes("claude-sonnet-5")) {
+    return {
+      maxOutputTokens: 128e3,
+      supportsStructuredOutput: true,
+      rejectsSamplingParameters: true,
+      rejectsThinkingDisabledAboveHighEffort: false,
       isKnownModel: true
     };
   } else if (modelId.includes("claude-sonnet-4-6") || modelId.includes("claude-opus-4-6")) {
@@ -3615,6 +3706,7 @@ function getModelCapabilities(modelId) {
       maxOutputTokens: 128e3,
       supportsStructuredOutput: true,
       rejectsSamplingParameters: false,
+      rejectsThinkingDisabledAboveHighEffort: false,
       isKnownModel: true
     };
   } else if (modelId.includes("claude-sonnet-4-5") || modelId.includes("claude-opus-4-5") || modelId.includes("claude-haiku-4-5")) {
@@ -3622,6 +3714,7 @@ function getModelCapabilities(modelId) {
       maxOutputTokens: 64e3,
       supportsStructuredOutput: true,
       rejectsSamplingParameters: false,
+      rejectsThinkingDisabledAboveHighEffort: false,
       isKnownModel: true
     };
   } else if (modelId.includes("claude-opus-4-1")) {
@@ -3629,6 +3722,7 @@ function getModelCapabilities(modelId) {
       maxOutputTokens: 32e3,
       supportsStructuredOutput: true,
       rejectsSamplingParameters: false,
+      rejectsThinkingDisabledAboveHighEffort: false,
       isKnownModel: true
     };
   } else if (modelId.includes("claude-sonnet-4-") || modelId.includes("claude-3-7-sonnet")) {
@@ -3636,6 +3730,7 @@ function getModelCapabilities(modelId) {
       maxOutputTokens: 64e3,
       supportsStructuredOutput: false,
       rejectsSamplingParameters: false,
+      rejectsThinkingDisabledAboveHighEffort: false,
       isKnownModel: true
     };
   } else if (modelId.includes("claude-opus-4-")) {
@@ -3643,27 +3738,39 @@ function getModelCapabilities(modelId) {
       maxOutputTokens: 32e3,
       supportsStructuredOutput: false,
       rejectsSamplingParameters: false,
+      rejectsThinkingDisabledAboveHighEffort: false,
       isKnownModel: true
     };
-  } else if (modelId.includes("claude-3-5-haiku")) {
+  } else if (modelId.includes("claude-3-haiku")) {
     return {
-      maxOutputTokens: 8192,
+      maxOutputTokens: 4096,
       supportsStructuredOutput: false,
       rejectsSamplingParameters: false,
+      rejectsThinkingDisabledAboveHighEffort: false,
       isKnownModel: true
     };
-  } else if (modelId.includes("claude-3-haiku")) {
+  } else if (/claude-(?:instant(?:-|$)|v?2(?=$|[-.:])|3(?=$|[-.]))/.test(modelId)) {
     return {
       maxOutputTokens: 4096,
       supportsStructuredOutput: false,
       rejectsSamplingParameters: false,
-      isKnownModel: true
+      rejectsThinkingDisabledAboveHighEffort: false,
+      isKnownModel: false
+    };
+  } else if (modelId.includes("claude-")) {
+    return {
+      maxOutputTokens: 128e3,
+      supportsStructuredOutput: true,
+      rejectsSamplingParameters: true,
+      rejectsThinkingDisabledAboveHighEffort: true,
+      isKnownModel: false
     };
   } else {
     return {
       maxOutputTokens: 4096,
       supportsStructuredOutput: false,
       rejectsSamplingParameters: false,
+      rejectsThinkingDisabledAboveHighEffort: false,
       isKnownModel: false
     };
   }