@ai-sdk/anthropic 3.0.80 → 3.0.82

package/src/anthropic-messages-options.ts

@@ -18,6 +18,8 @@ export type AnthropicMessagesModelId =
   | 'claude-sonnet-4-6'
   | 'claude-opus-4-6'
   | 'claude-opus-4-7'
+  | 'claude-opus-4-8'
+  | 'claude-fable-5'
   | (string & {});
 
 /**
@@ -223,6 +225,34 @@ export const anthropicLanguageModelOptions = z.object({
    */
   inferenceGeo: z.enum(['us', 'global']).optional(),
 
+  /**
+   * Server-side fallback chain.
+   *
+   * 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.
+   *
+   * The required `server-side-fallback-2026-06-01` beta is added automatically
+   * when this option is set.
+   */
+  fallbacks: z
+    .array(
+      z.object({
+        model: z.string(),
+        max_tokens: z.number().int().optional(),
+        thinking: z.record(z.string(), z.unknown()).optional(),
+        output_config: z.record(z.string(), z.unknown()).optional(),
+        speed: z.enum(['fast', 'standard']).optional(),
+      }),
+    )
+    .optional(),
+
   /**
    * A set of beta features to enable.
    * Allow a provider to receive the full `betas` set if it needs it.

package/src/anthropic-tools.ts

@@ -42,7 +42,7 @@ export const anthropicTools = {
    * Supported executor models: Claude Haiku 4.5, Sonnet 4.6, Opus 4.6,
    * Opus 4.7. The advisor must be at least as capable as the executor.
    *
-   * @param model - The advisor model ID (required), e.g. `"claude-opus-4-7"`.
+   * @param model - The advisor model ID (required), e.g. `"claude-opus-4-8"`.
    * @param maxUses - Maximum advisor calls per request (per-request cap).
    * @param caching - Enables prompt caching for the advisor's transcript
    * across calls within a conversation. Worthwhile from ~3 advisor calls

package/src/convert-anthropic-messages-usage.ts

@@ -5,26 +5,25 @@ import type { JSONObject, LanguageModelV3Usage } from '@ai-sdk/provider';
  *
  * - `compaction` / `message`: executor iterations, billed at executor rates.
  * - `advisor_message`: advisor sub-inference, billed at the advisor model's
- *   rates. The `model` field carries the advisor model ID. Advisor tokens
- *   are NOT rolled into the top-level totals because they bill at a
- *   different rate; inspect this array for advisor cost tracking.
+ *   rates. Advisor tokens are NOT rolled into the top-level totals because
+ *   they bill at a different rate; inspect this array for advisor cost
+ *   tracking.
+ * - `fallback_message`: a server-side fallback attempt that served the turn.
+ *   When present, the top-level usage already reflects the served answer, so
+ *   it is used as-is.
+ *
+ * The `model` field carries the model that produced the iteration. The API
+ * populates it for the per-model attribution cases (the fallback chain and
+ * advisor sub-inferences) and omits it otherwise.
  */
-export type AnthropicUsageIteration =
-  | {
-      type: 'compaction' | 'message';
-      input_tokens: number;
-      output_tokens: number;
-      cache_creation_input_tokens?: number | null;
-      cache_read_input_tokens?: number | null;
-    }
-  | {
-      type: 'advisor_message';
-      model: string;
-      input_tokens: number;
-      output_tokens: number;
-      cache_creation_input_tokens?: number | null;
-      cache_read_input_tokens?: number | null;
-    };
+export type AnthropicUsageIteration = {
+  type: 'compaction' | 'message' | 'advisor_message' | 'fallback_message';
+  model?: string | null;
+  input_tokens: number;
+  output_tokens: number;
+  cache_creation_input_tokens?: number | null;
+  cache_read_input_tokens?: number | null;
+};
 
 export type AnthropicMessagesUsage = {
   input_tokens: number;
@@ -57,10 +56,19 @@ export function convertAnthropicMessagesUsage({
   // and output_tokens exclude compaction usage. Advisor (`advisor_message`)
   // iterations are filtered out: they bill at the advisor model's rates,
   // not the executor's, so they don't belong in the top-level totals.
+  //
+  // A turn served by a server-side fallback is the exception: the served
+  // answer comes from the fallback model, so the executor `message` iteration
+  // is the blocked primary attempt (zero output). The top-level totals already
+  // reflect the fallback answer, so they are used directly.
   let inputTokens: number;
   let outputTokens: number;
 
-  if (usage.iterations && usage.iterations.length > 0) {
+  const servedByFallback = usage.iterations?.some(
+    iter => iter.type === 'fallback_message',
+  );
+
+  if (usage.iterations && usage.iterations.length > 0 && !servedByFallback) {
     const executorIterations = usage.iterations.filter(
       iter => iter.type === 'compaction' || iter.type === 'message',
     );

package/src/convert-to-anthropic-messages-prompt.ts

@@ -157,15 +157,8 @@ export async function convertToAnthropicMessagesPrompt({
 
     switch (type) {
       case 'system': {
-        if (system != null) {
-          throw new UnsupportedFunctionalityError({
-            functionality:
-              'Multiple system messages that are separated by user/assistant messages',
-          });
-        }
-
-        system = block.messages.map(({ content, providerOptions }) => ({
-          type: 'text',
+        const content = block.messages.map(({ content, providerOptions }) => ({
+          type: 'text' as const,
           text: content,
           cache_control: validator.getCacheControl(providerOptions, {
             type: 'system message',
@@ -173,6 +166,13 @@ export async function convertToAnthropicMessagesPrompt({
           }),
         }));
 
+        if (system == null) {
+          system = content;
+        } else {
+          messages.push({ role: 'system', content });
+          betas.add('mid-conversation-system-2026-04-07');
+        }
+
         break;
       }
 

package/src/tool/advisor_20260301.ts

@@ -77,7 +77,7 @@ const factory = createProviderToolFactoryWithOutputSchema<
     },
   {
     /**
-     * The advisor model ID, such as `"claude-opus-4-7"`. Billed at this
+     * The advisor model ID, such as `"claude-opus-4-8"`. Billed at this
      * model's rates for the sub-inference.
      *
      * The advisor must be at least as capable as the executor; an invalid